In a small room at the Broad Institute, a glass vessel the size of a coffee thermos has been running for eleven days. A clear medium drips into it through one tube; a thinner stream of waste flows out through another. Inside, the population of E. coli doubles every twenty minutes or so, and every twenty minutes most of those cells are washed away. The cells that stay are the ones carrying a virus — a bacteriophage called M13 — that has mutated, just enough, to keep doing the trick the experiment demands.
This is phage-assisted continuous evolution, PACE for short, and the man who built it is David Liu, a chemist who decided in the mid-2000s that the bottleneck in directed evolution wasn’t biology. It was patience.
To understand why, it helps to know what directed evolution had been before. The technique itself goes back to Frances Arnold’s work at Caltech in the 1990s, work that earned her the Nobel Prize in 2018. The recipe is simple in outline: take a gene encoding a protein you’d like to be better at something — splitting a stubborn chemical bond, say, or recognizing a new molecular target. Make many copies, each with random mutations. Express them in cells. Screen each variant. Pick the best one. Mutate again. Repeat.
It works. It made enzymes that operate in solvents Arnold’s contemporaries thought would denature any protein. It built antibody libraries that drug companies still mine today. It is also, in its classical form, slow. Each round of mutation and selection took a week or more — pipetting, plating, screening, sequencing. To evolve a protein through a hundred generations of selection was a year of someone’s life.
Liu’s bet was that biology already knew how to do this faster. Viruses evolve in hours, not weeks, because they have no patience either. Their selection pressure is whether they get to make the next generation. If you could couple a protein’s function to a virus’s ability to replicate, evolution would run on viral time.
The architecture of PACE is built around that coupling. The M13 phage has a gene, gIII, that it needs to make infectious copies of itself. Liu’s group stripped gIII out of the phage’s own genome and put it on a plasmid in the host E. coli — but they wired it so the host only expresses gIII when the protein under evolution is doing what the experimenter wants. A phage carrying a successful protein variant gets gIII expressed in its host, makes infectious progeny, propagates. A phage with a defective variant infects a cell, finds no gIII, and produces non-infectious dead ends.
The “continuous” in continuous evolution refers to the bioreactor. The vessel — Liu’s group calls it a “lagoon” — is kept in a steady state by constant inflow of fresh host cells and outflow of medium. The phage population must reproduce faster than it is washed out, or it goes extinct. Selection is unrelenting because dilution is unrelenting.
The numbers are bracing. A PACE run can move through a hundred rounds of selection in eight days, the equivalent of two human-years of classical directed evolution. It has produced new TALENs, polymerases that read modified bases, base editors with rewritten substrate preferences, and a steady output of papers that read like dispatches from a faster planet.
The lesson PACE points to is older than the apparatus. When you cannot design the thing you want, you can sometimes build a world in which the thing you want is what survives. The trick is making the world cheap and the survival meaningful. Liu built one such world in glass.
The natural follow-up is to look at base editors and prime editors — Liu’s other line, which combines CRISPR’s specificity with PACE-evolved enzymes to make single-base edits without cutting DNA. The 2017 (CBE) and 2019 (prime editing) papers are foundational; the most recent work on therapeutic delivery is where the field’s energy is now.