How David Liu turned Botox into his next drug discovery platform — a molecular axe that can maybe shred Tau, RAS and a host of the worst disease-causing proteins

Before David Liu became famous for inventing new forms of gene editing, he was known around academia in part for a more obscure innovation: a Rube Goldberg-esque system that uses bacteria-infecting viruses to take one protein and turn it into another.

Since 2011, Liu’s lab has used the system, called PACE, to dream up fantastical new proteins: DNA base editors far more powerful than the original; more versatile forms of the gene editor Cas9; insecticides that kill insecticide-resistant bugs; enzymes that slide synthetic amino acids into living organisms. But they struggled throughout to master one of the most common and powerful proteins in the biological world: proteases, a set of Swiss army knife enzymes that cut, cleave or shred other proteins in everything from viruses to humans.

Despite their prowess, proteases have been used as drugs only a handful of times in history. But Liu thought that, with some heavy tuning, they could let researchers do to proteins what the first gene editors let them do to DNA. Every protease binds and cuts a specific string of amino acids, the building blocks of proteins. In other words, they cut whenever and wherever they see a code. If you could engineer proteases to go after a different code — the code for a protein that drives cancer, say, or the tangles that build up in patients with Alzheimer’s — you could cut those proteins out of patients. You might even turn a misshapen protein back into a healthy one.

“If you think of virtually any biological process, including virtually any disease, there are proteins that, if you imagine cleaving, would probably disrupt that process or change that process,” says Liu, a professor at the Broad Institute and Harvard and co-founder of the gene editing companies Beam, Editas and Prime. “It’s not difficult to imagine how being able to cleave proteins at specified sequences might provide many new opportunities.”

Late last month, in Science, Liu and a postdoc, Travis Blum, provided the first proof-of-concept for what they call proteome editing. They took a protease already used in humans and, by simulating millions of years of evolution, taught it to precisely cut a protein that looks drastically different than its natural target.

The paper builds on the work of researchers who have labored for decades to reprogram proteases into drugs for various diseases. They praised the new effort.

“I think it’s a big advance,” says Grant Blouse, who oversees research for Catalyst Biosciences, one of the only biotechs focused on protease drugs. “There’s a lot of potential.”

Liu cautioned that the technology is still in its early stages. He compared it to the invention of zinc-finger nucleases, one of the first gene editing technologies, in the ’90s, more than two decades before anyone would actually edit genes in humans. He will have to show the same system can manipulate other proteases to slash other proteins and that they can do so in humans without triggering too powerful of an immune response.

Still, outside scientists are already beginning to imagine the possibilities for tripping up pathogenic proteins in diseases like Parkinson’s and shredding cancer genes developers have long struggled to inhibit, such as KRAS. More immediately it could be used as a research tool, allowing biologists to study what happens when you eliminate a protein at different stages of an organism’s life cycle.

“There is no limit to what you can design” with Liu’s system, says Cesare Montecucco, a pathologist at Italy’s University of Padua who has studied proteases for four decades. “You are in a room with many, many doors.”

The true beauty of Botox

Liu chose to conduct his proof-of-concept study on the world’s most famous protease: Botox. Most Americans would probably not rank Allergan’s billion-dollar beauty molecule among the last century’s greatest medical innovations, but the protease has long fascinated biologists.

A product of one bacterium’s eons-long quest to control its host animals, botulinum toxin, often abbreviated as BoNT, is one of the most potent drugs in existence. Just a few billionths of a gram can sneak into neurons and have an effect for months, slicing the synapses that allow muscles to contract and skin to wrinkle. The dose is small enough to fly under the radar of an immune system that usually prowls for foreign proteins. Those properties have led doctors to prescribe Botox for everything from depression to drooling.

“It’s kind of the perfect protein drug,” says Min Dong, who runs a BoNT lab at Harvard and Boston’s Children Hospital and co-authored Liu’s paper. “I’ve been studying them for 20 years. I’m still fascinated.”

BoNTs cut a protein called SNARE. For years, Dong and a handful of other researchers tried to tweak the toxin to cut related proteins or enter neurons with greater fluidity. They used their knowledge of its structure or introduced random mutations until they found beneficial ones. Others applied the same technique to proteases the body relies on for clotting blood, like the hemophilia drug Factor IX.

Those techniques, though, didn’t allow scientists to radically alter proteins. That’s where Liu stepped in. PACE, short for phage-assisted continuous evolution, effectively automates the random mutation system most researchers used, allowing a given PhD student to go from engineering a few proteins in a 5-year program to engineering 50 to 100.

“It enabled us to go after targets, protein evolution goals that were much more ambitious that we would have dared to try using conventional methods,” Liu said.

The process begins with a virus that infects bacteria, called a phage, and the bacteria E. coli. Researchers took out a gene the phage needs to replicate and replaced it with a gene for botulinum toxin. The E. coli is engineered in two ways: It has a “mutagenic plasmid” that makes the phage mutate faster than it usually would, and it has the phage’s missing gene. But the gene is locked up, shackled behind a protein that looks a bit different than what BoNTs usually cut. Only phages that express the right mutant BoNT can cut it open and replicate.

Over successive cycles, researchers gradually shift the lock to look less and less like the original protein and more and more like the protein they want to target. It’s as if you wanted to evolve very tall monkeys, so over successive generations, you moved branches with bananas higher and higher.

That’s the easy part, though. Liu already showed he could evolve proteases from hepatitis C and a tobacco virus to cut far-flung proteins, but they would never work as drugs. Those proteases cut a new target, but they still shredded plenty of other ones as well. You might as well swallow a power drill.

So Blum, Liu’s postdoc, designed a “negative selection system.” Essentially, he added a second lock, this one containing a poison, that BoNTs would naturally open, killing the phage. Only proteases that couldn’t cut the lock survive. By mixing both systems, they could develop proteases that cut the target and none of the ones it naturally pursued.

“It’s really a tour de force of a technical advance,” says Bryan Dickinson, who runs a protein engineering lab at the University of Chicago. “It’s the combination that makes this work.”

Liu and Blum evolved their BoNT over several hundred hours to cut PTEN, a protein that restricts cell growth and looks almost nothing like SNARE. They asked to meet with Dong, who looked over the results and told them he thought that would be impossible. “It was a total surprise,” he says.

‘A class of its own’

PTEN-cutting Botox could itself be a major accomplishment. Neuroscientists long theorized that knocking out PTEN could help neurons regrow, but they worried knocking it out across the body could lead to unwanted cell growth, i.e. cancer. Dong and Liu are now testing whether their BoNT, which only enters neurons, can help treat stroke or spinal cord injury by triggering nerve regeneration.

Carolyn Bertozzi, a Stanford chemist who studies protein degradation, says one of the key assets is that the BoNTs can go inside neurons. Many drugs, including antibodies, can only reach proteins on the surface and between cells. You could in theory use PACE to make a BoNT that shreds tau, the tangles that accumulate inside and ultimately spill out of brain cells in Alzheimer’s patients, allowing you to target the disease earlier and more thoroughly than other approaches allow.

The biggest question is precisely how programmable proteases are. Can you engineer BoNTs to cut any protein? Probably not, Liu says. But there are other proteases you could start with, from humans or bacteria or viruses, and there’s no obvious limit to the targets you could pursue. Those other proteases may also let you get into other organs, such as the liver or gut.

“Surely, it won’t be fully programmable, but the more programmable we can get it to, the better off we are,” says Chang Liu, who runs a protein engineering lab at UC Irvine. “It is still, of course, unknown how far they can get.”

It’s easy, though, to speculate about targets: RAS and MYC, two common cancer-fueling proteins that have been making fools of drugmakers for decades, for example; or transcription factors, regulatory proteins implicated in a host of diseases but which have long been nearly impossible to hit.

Investors have poured billions into the protein degradation companies over the last few years in an effort to hit some of these same targets, seeking to develop molecules that connect harmful proteins to the body’s internal garbage disposal system (which is made up of proteases).

But Bertozzi says Liu’s approach has distinct advantages. Degrading companies still need to find molecules that can latch onto the smooth surfaces of RAS or MYC, which they’ve struggled to do. And degraders are a blunt tool: They only destroy.

“It falls in a way into a class on its own,” Bertozzi says. “It’s actually quite a bit more versatile. I think that this technology has the potential to do more than just disable.”

In theory, proteases could clip apart “fusion proteins,” such as BRCA1, that drive cancer. Or trim the abnormally long protein that causes Huntington’s disease back down to size. Those applications will be harder, though, says Matt Bogyo, a professor of pathology and a chemical biologist at Stanford.

“Proteases permanently alter the protein, they clip it into fragments,” Bogyo says. “Most of what you’re doing is destroying.”

That pesky immune system

To ever work as a drug, Liu’s proteases face a key hurdle: the human immune system.

Charles Craik, a chemist at UCSF and a co-founder of Catalyst, was one of the first scientists to try to broadly engineer proteases as drugs in the ’90s. He praised Liu’s work as a major step forward in protein engineering, but he questioned how translatable it would be.

Botox sneaks under the immune system because it’s given in trace doses, but many of the potential applications would require higher amounts. One solution would be to develop proteases from humans instead of bacteria, but Liu’s platform would then take that native protease and scramble it into what may look to immune cells as an exotic and potentially dangerous object.

“You can make a few changes, but you can’t make a lot of changes, because then it’s really different,” Craik says. “I’m not saying this technology couldn’t be applied to therapeutic enzymes, but now you’re going to have to deal with as few mutations as possible to keep it the same.”

As antibodies began taking over biotech in the ’90s, Craik thought that proteases could be used at a similar scale. PACE could offer a platform to do that, analogous to the technologies Regeneron and Genentech developed to extract antibodies from mice and humans.

It’s not a simple platform, though. Although David Liu first published the technology a decade ago, it is still not widely used, researchers said, in part because it requires significant expertise — a hurdle that could prevent more labs and companies from using it to develop proteases for various targets.

“There are a lot of moving parts,” says Chang Liu.

Back in the Liu lab

Liu, of course, knows the platform well. He says he hasn’t drawn out a formal list on a whiteboard or anything, but he admitted what other researchers, knowing his past exploits, guessed: He’s already gaming out a long list of potential applications.

It’s too early, Liu says, to get into specifics, but he laid out a few criteria. They would go after areas where giving proteases just once might have long-term effects: for example, on proteins that control how cells regenerate and differentiate or hormones that control a cascade of complex processes.

It could include areas where they would want to activate or change a protein rather than simply destroy it, he says. There’s almost a surfeit of possibilities.

“Virtually every disease or biological process involves proteins, so it’s not difficult to imagine [applications],” Liu says. “Where we want to be wise in selecting targets is to make sure that we’re addressing problems that are difficult to address using other approaches, or people have tried and not had much success.”