PHILADELPHIA — Cesar de la Fuente believes the next breakthrough antibiotic might come from animals that have been dead for thousands of years.
Since 2021, his lab here at the University of Pennsylvania has built algorithms to trawl genetic databases for protein fragments, called peptides, with microbe-squashing properties. They started with human DNA. But more recently, he’s looked deep into the fossilized past, hunting for potential drugs lurking in the code of Neanderthals, giant sloths and woolly mammoths, among other ancient animals.
His team uses robots to resurrect the most promising snippets and then tests whether they can clear infections in mice at rates comparable to the standard antibiotic polymyxin B. Last year, he named the approach “molecular de-extinction” — a much safer, more feasible and perhaps less lucrative version of Jurassic Park.
“What if we brought back molecules instead of just an entire organism?” he said.
It is, suffice to say, an unorthodox approach to tackling the growing threat of antibiotic resistance, which it’s estimated killed 1.2 million people in 2019. Other researchers try to map a bacterial cell’s precise vulnerabilities, or the weapons they’ve evolved to neutralize existing drugs, then find new ways to jam them.
“Essentially, it’s a time machine,” said Kim Lewis, who develops new antibiotics at Northeastern University and is not involved in the work.
De la Fuente’s research, detailed in a July publication and preprint earlier this month, relies on genomes painstakingly cobbled together by paleogeneticists over decades: Fragments of A, T, G, and Cs fished out of ancient bones, tundra-buried carcasses, and 19th century taxidermy and strung together into a coherent text.
Those scientists were interested in understanding evolution, work that won Svante Pääbo a Nobel Prize last year, but the sequences remained freely available online. De la Fuente saw a potential medical trove.
“I looked at the papers and they looked reasonable,” said Nick Patterson, a Broad Institute geneticist who helped decode the Neanderthal and Denisovan genomes, before adding a question. “What about modern mammals?… I’m sure there’s an enormous amount out there in the modern mammals we don’t know about.”
De la Fuente says modern mammals are next, along with most known viruses and most known bacteria. Some researchers, including Lewis, are skeptical that a giant elk or sea cow peptide will ever cure a human infection, noting the previous failure of similar drugs. But De la Fuente sees the research as part of a broader case for how scientists might use our ever-expanding genetic libraries to find new drugs against infections and other diseases.
“Basically,” he said, “our goal is to mine everything to see how many antimicrobial sequences we can find.”
The idea of resurrecting ancient protein shards emerged in a lab meeting a couple of years ago. By then, de la Fuente was already known for quirky work that bridged pipette-and-beaker microbiology with new machine learning techniques.
It was a ready fit. He grew up in the Spanish port city of A Coruña, where he was fascinated enough by nature to collect marine organisms off the beach and dissect them in his home, and nerd enough to ask Santa for the exact number of helium balloons he calculated would be needed for a flying chair.
Researchers had long borrowed antibiotics from nature. Penicillin was discovered in a soil fungus, as were other stars of the golden age of antibiotic discovery. But after the 1970s, researchers struggled to find new molecules, having seemingly tapped most of what the microbial communities beneath our feet had to offer.
De la Fuente believed computers could break the bottleneck. After a Ph.D. in antibiotic discovery at the University of British Columbia, he landed at MIT, “the mecca for AI.”
“But people were using it mostly for pattern recognition and speech recognition algorithms, used for other things, not for biology or medicine,” he said.
His early work improved on nature’s wimpier molecules. He’d take a peptide in guava seeds or wasp venom already known to have activity against bacteria and design an algorithm to generate thousands of variants before selecting the ones likely to be the most potent — evolution on a chip.
But he and his team also built algorithms that acted as a magnifying glass, searching for new sequences that might have previously undiscovered bacteria-killing properties. In 2021, they looked at the entire human proteome — every portion of DNA known to code for a protein — and found over 2,000 such peptides.
Many were surprising. Biology textbooks say that genes code for proteins. Really, though, genes code for peptides. Some peptides, such as insulin, function on their own. Most act like machine parts, folding with other peptides into proteins that do the real work.
De la Fuente’s paper showed that many of those machine parts had antimicrobial properties, even if the machine itself was dedicated to, say, keeping neurons alive or the heart pumping. What if it wasn’t just the immune system that fought off infections, but “encrypted” components of wholly unrelated proteins?
“I suspect that these things have been evolving in humans probably since we’ve been around,” said Jim Collins, an MIT synthetic biologist who has worked with de la Fuente, calling it “background defenses.”
They started looking at human ancestors: Neanderthals and the lesser known Denisovans, whose genomes were deciphered around a decade ago. That work raised new questions, some they contemplated from the outset. Were there any ethical concerns with reviving an extinct molecule? Probably not, but best that they keep the peptides in the freezer. Could such molecules be patented? Normally, you can’t patent molecules found in nature, but these didn’t exist in nature anymore.
Their search found numerous more, including in Neanderthal ATP synthase, a core turbine for creating cellular energy.
De la Fuente keeps a plastic model of the protein in his office. It was an amorphous tangle of gray helixes, with a single red helix on top. He detached the red spot.
“This is an encrypted sequence that we call neanderthalin-1,” he said. Lab robots had recreated the actual version in a dish and students gave it to mice. “This actually was active in our preclinical animal model.”
De la Fuente insists his work has practical relevance, but he admits he was also fascinated by its Crichton-esque nature. He looked teenagery in blue jeans, white T-shirt and denim jacket. The Princeton Field Guide to Prehistoric Mammals sat in a large cardboard box, showing what appeared to be a giant sloth mauled by two saber tooths.
He acknowledged neanderthalin-1 isn’t potent enough to be a good antibiotic. But after Neanderthals, they searched what he called — because biologists can rarely resist an “ome” — the “extinctome”: The 208 extinct organisms with sequenced genomes available online. They found thousands and synthesized 69 of the most intriguing, including 20 not known in modern animals.
These proved more promising. Two performed similarly to a leading antibiotic on both mouse skin and deep thigh infections — a peptide from an ancient elephant and a peptide from mylodon, a cave-dwelling, nearly 10-foot long sloth that vanished around 12,000 years ago.
“I find it fascinating — it was initially found in Patagonia and Argentina, I think by Charles Darwin in one of his adventures there,” he said.
Peptides, however, have a sordid history as antibiotics. They once generated significant excitement but many proved too toxic because they target membranes, a core component of both bacterial and human cells.
“Efforts over the past decades to develop this class of compounds into systemically acting antibiotics — it has not yet been successful,” said Lewis, the Northeastern professor.
Still, Daria Van Tyne, who researches antibiotic resistance at the University of Pittsburgh, said she was “pleasantly surprised,” by the mouse data. With infections rising globally and antibiotic companies facing their own extinction worries, this was the kind of out-of-the-box thinking the field needed.
“We are in a position where we have to get creative,” she said. “The future looks pretty bleak.”
It’s also not clear why ancient peptides would necessarily be better than ones that exist in nature today, or ones built in a lab. George Church, the Harvard geneticist trying to resurrect the actual mammoth — tusks and all — said de la Fuente could repurpose his algorithm to design new peptides from scratch.
De la Fuente said the technology isn’t quite there yet, but he shares the ultimate goal of designing peptides from scratch. By mining these vast sequences, his lab has been able to make more and more sophisticated algorithms to identify and create antibiotics.
In the human proteome work, they specified the exact properties, such as electric charge and sequence length that they believed mark a peptide. But in the ancient genome work, they used machine learning — the same technique deployed in AlphaFold and ChatGPT — to find more subtle qualities humans may have overlooked.
He’s hoping to find a way to advance his peptides toward the clinic, a difficult task given the broken market for novel antibiotics, while continuing to work his way through the tree of life. They’re mainly held up by computer power. A postdoc recently calculated they would need to run their computers nonstop for 500 days to mine every known sequence.
“So it’s not feasible for us,” he said, before floating a possible pitch. “I’m sure if Google would lend us some of their computer power, we could do it.”