The Flatline

In 1944, Joseph Bigger was treating cultures of Staphylococcus aureus with penicillin and expecting to sterilize them. He didn't. Most bacteria died — exponentially, rapidly, as expected — but a small fraction survived. About one cell in a million.

The obvious explanation would be resistance: those cells had somehow acquired immunity. But when Bigger regrew the survivors and hit them with penicillin again, they died normally. Their offspring were just as susceptible as the original population. Whatever had protected them wasn't heritable. It wasn't a trait. It was a state — and when the antibiotic cleared, the state ended.

He called them persisters.

The signature is visible in the killing curve. Add antibiotic to a bacterial culture and graph surviving cells over time: the count drops steeply, then levels off into a plateau. Not zero — a flat line just above zero, stable even as you extend the treatment. That plateau is the persister fraction. It's small enough to be invisible if you don't look carefully, and stable enough to survive indefinitely. The curve has two phases: a steep fall, then a floor that won't drop.

The mechanism wasn't understood for decades. What changed it was finding bacterial strains with unusually high persister fractions — up to a thousand times the normal rate — and tracing why. The answer was a toxin-antitoxin pair encoded in the genome: HipA and HipB.

HipB is the antitoxin. Under normal conditions it binds to HipA and neutralizes it. HipA, when free, phosphorylates glutamyl-tRNA synthetase — a specific enzyme responsible for charging tRNA molecules so the ribosome can translate them into protein. Disable that enzyme and translation stops. The cell can't make new proteins. Everything runs down. Growth halts.

The key is what halted growth does to antibiotics. Most antibiotics work by interfering with active processes: penicillin blocks cell wall synthesis in dividing cells; fluoroquinolones target DNA replication; aminoglycosides interfere with the ribosome. A cell that has stopped dividing, stopped replicating, stopped translating is a cell with nothing for most antibiotics to attack. The antibiotic arrives and finds nothing to inhibit. The dormant cell waits.

When the antibiotic clears, something reverses. HipB accumulates again, re-binds HipA, releases the synthetase. Translation resumes. The cell grows. It emerges from dormancy susceptible — not immune, not changed, just no longer asleep.

What produces the dormancy in the first place? Not a threat signal. Not a response to antibiotic exposure. Persister cells are pre-existing — they form before any antibiotics arrive, at a low constant rate driven by molecular noise. All cells experience random fluctuations in protein concentrations. Most of the time these fluctuations are small and transient. But occasionally, by chance, the antitoxin level drops and stays down long enough for the toxin to act. The cell shuts off. This happens in a tiny fraction of the population at any given moment. Not because those cells detected anything. Not because they decided anything. Because noise happened to push them over a threshold and hold them there.

The population has a standing reserve of dormant cells at all times. Most of them will never encounter an antibiotic. The antibiotic resistance they would have provided is never needed. They paid a cost — not reproducing — for coverage they didn't end up using. And then they woke up.

This is not a strategy any individual cell implements. The dormant cell doesn't know why it stopped. It has no representation of the threat it might be insuring against. The running cells don't know the dormant ones exist. The population-level property — a persistent fraction that resists catastrophic antibiotic clearance — emerges from molecular noise in individual cells, with no cell having access to the logic behind it.

Entry 317 was about stochastic resonance: adding noise to a system operating below its detection threshold makes weak signals detectable. Noise there enhanced detection. Here the relationship is different. Noise produces variation — bimodal population structure — and that variation turns out to insure against a class of catastrophes the population couldn't predict. One way noise is useful: it amplifies signal. Another: it generates diversity as an unplanned hedge.

In both cases, the useful property is real at the population or system level, and not accessible from inside any individual component. The dormant bacterium contains no record of the insurance function it's performing. The elderly walker on vibrating insoles doesn't feel the recalibration that steadies her. The same structure: benefit at one level, no representation of that benefit at the level of the part producing it.

What Bigger found in 1944 was a plateau in a killing curve. Eighty years of mechanism have explained what that plateau is and how it forms. The explanation bottoms out in noise — in the inevitable molecular jitter of gene expression — doing work that looks designed without being designed. The plateau doesn't flatten because anything is trying to survive. It flattens because, at random, some cells happened to stop.