Scientists Learn to Both Build New Proteins and Erase Old Ones

A protein is really just a long word, spelled out letter by letter from a twenty-letter alphabet that life settled on billions of years ago. Every antibody, every enzyme, every hormone in a human cell is built from that same fixed vocabulary. This summer, two teams of scientists effectively rewrote the dictionary from both ends: one, at Harvard Medical School, worked out how to write brand-new letters into that alphabet inside a test tube, while the other, at Weill Cornell Medicine, built a calculator that tells drugmakers exactly how to erase a word once a protein turns dangerous. Between them, the two advances make protein engineering a two-way street, as good at construction as it is at demolition.

Neither team set out to answer the other’s question. But drop their results side by side and a shared ambition appears: control over proteins at the molecular level, whether the goal is inventing one that has never existed or dismantling one that is already causing harm.

The construction side comes from a tool called AGENTEX, short for automated genetic tRNA expansion, built in George Church‘s lab. Cells build proteins from just 20 amino acids, even though the genetic code has 64 possible three-letter codons to work with, a redundancy that has tempted genetic engineers for years since it hints some codons are free to be repurposed. The Church lab spent years proving in 2013 that a single codon could be freed up in living E. coli bacteria and reassigned to a new, human-made amino acid. It then took another decade to free a second one, bringing the total to 22 amino acids per protein. AGENTEX compresses that timeline dramatically: using a library of 48 different tRNA sequences and a cell-free soup of E. coli components, the team built protein-making systems that can call on up to 34 amino acids at once, all without touching a living genome.

Rewriting the Rules of the Ribosome

Getting there required overturning a rule that had stood since the earliest days of molecular biology. Every tRNA molecule, the shuttle that ferries an amino acid to the ribosome for assembly, ends in the same three-letter sequence, CCA (short for its RNA letters: cytosine, cytosine and adenine). Textbooks held that any other ending would be rejected outright, both by the enzymes that load the amino acid and by the ribosome that reads it. Felix Radford, the study’s first author, found that the textbook rule was only partly right. “We’ve shown that we can alter one of the most fundamental portions of one of the most fundamental systems found in nature, the protein-synthesis system that has existed pretty much unchanged for billions of years across all organisms, and it’s functional. The CCA end is much more flexible than people assumed,” Radford says. The loading enzymes, it turns out, will still hand an amino acid to a tRNA with a swapped ending. Only the ribosome still insists on the original tag, which the team exploited by engineering a matched ribosome that recognizes the new ending instead, keeping the synthetic system quarantined from an organism’s own machinery. In one test, a ribosome built to work with the modified tag produced about six times more of a target protein than an unmodified one did, and roughly ten times more than a reaction with no genetic template at all.

“AGENTEX enables researchers to generate entirely new genetic codes on demand in test tubes and use them at scale to build proteins far beyond what nature has evolved,” Radford says. “This is more rapid and safe than existing methods, as it does not rely on handling living cells or altering their genomes.” Church, the paper’s senior author, put the leap in blunter terms: “It took us a decade per new amino acid added to the code, so this remarkably opens the door to 34 at once and with almost none of the usual collateral damage to the genome.”

A Calculator for Taking Proteins Apart

A thousand miles south, at Weill Cornell Medicine, Wei Du and Olivier Elemento have been working the opposite end of the same problem: not how to build a protein, but how to make one disappear. A newer class of cancer drugs, called protein degraders, does not simply block a disease-causing protein the way conventional drugs do. It hijacks the cell’s own disposal system and flags the protein for destruction outright, which matters when the problem is not a protein behaving badly but a protein existing in far too many copies.

Designing a degrader has mostly been trial and error, since a small change to the molecule’s connecting linker can be the difference between a drug that works and one that does nothing. Du built a mathematical model instead, one that predicts how efficiently a given target will be destroyed from measurements a lab can obtain relatively easily, rather than from years of synthesizing and testing candidate molecules. Trained against published data on 41 protein targets, the model settled on a striking pattern: the actual rate at which a marked protein gets broken down by the cell’s disposal machinery falls into a narrow physical range across almost every target tested, no matter how different the underlying biology.

The model held up outside its training set, too. Applied cold to a panel of 37 disease-linked kinases with no further tuning, its predictions landed within 15 percent of the measured outcome 81 percent of the time. In one direct laboratory check, the team dosed batches of 500,000 human cells with a candidate degrader across a range of concentrations and time points, and the model’s predictions tracked the measured results closely.

“We showed that for protein targets with slow turnover, even relatively weak degrader binding affinity can result in potent degradation,” Du says. “We also generated a list of potential targets with high value for degrader development.” That finding cuts against a long-standing drug-discovery instinct, which has treated a tight molecular grip as the main thing worth optimizing. For a protein that already breaks down slowly on its own, the model suggests a looser-binding, easier-to-develop degrader can still finish the job.

Elemento frames the model as one piece of a larger ambition that has little to do with efficiency for its own sake. “This is one step toward our goal, which is not to design a drug that is supposed to work for a million patients or a thousand patients, but to design one that is customized to one unique patient,” he says.

Neither AGENTEX nor the degradation model has been tested in a living organism or a patient, and both papers are candid about that ceiling. AGENTEX’s authors note it remains unclear what further engineering the system will need before it can operate inside cells rather than cell-free lysate. The degradation model, for its part, does not yet account for factors like competing proteins or feedback loops that could complicate its predictions inside an actual cell, and its target lists describe population-level opportunities rather than any single patient’s biology.

FAQ