Plasmid addiction systems

Why does a bacterium keep carrying genes it would plainly be better off dropping?

A plasmid is a small loop of DNA that sits alongside a bacterium's chromosome and is not part of it. Copying it costs the cell something: nucleotides, polymerase time, and the burden of whatever proteins it insists on expressing. Take away the antibiotic that makes its resistance gene worth having, and the plasmid is pure overhead.

So here is the puzzle. Segregation at cell division is not perfect. Sooner or later a daughter cell is born without a copy. That cell is now cheaper to run than its cousins, and cheaper means faster, and faster means it should take over the flask within a few dozen generations. Yet plasmids persist in populations for years without any selection for the genes they carry. What is holding them there?

The first instinct is that the plasmid must be smuggling some hidden benefit — and sometimes it is. But watch what happens if we push on the other side of the ledger instead. The plasmid does not have to make its host better. It only has to make not carrying it worse. Those are different problems, and the second is far easier to solve.

Suppose a plasmid encodes two things: a poison, and the antidote to that poison. While the plasmid is present, both are made, and they cancel. Nothing happens; the cell is unharmed. Now ask what the plasmid-free daughter inherits. She got no plasmid, so she can make neither protein again — but she did inherit whatever quantity of each was floating in the cytoplasm at the moment of division.

Which raises the question that the whole trick turns on: do the two molecules last equally long? They do not, and the asymmetry is deliberate. The toxin is a stable protein. The antitoxin is fragile — either a protein with a degradation tag that host proteases such as Lon or ClpXP chew through in minutes, or, in another family, a small antisense RNA with a very short half-life that works by blocking the toxin's message rather than the toxin itself. In a cell with the plasmid, that fragility is invisible, because fresh antitoxin is being made continuously. In a cell without it, the antidote drains away while the poison sits there intact.

Notice what this does to the accounting. The plasmid-free daughter did not merely fail to gain something. She was killed or arrested by an inheritance she could not refuse. The label the field uses is exact: post-segregational killing. The whole apparatus is a commitment device, written in chemistry. The plasmid has arranged the payoffs in advance so that the option of leaving is removed from the host, not argued out of it.

And a commitment device only works if the threat cannot be revoked once made. That is why the timing asymmetry matters more than the potency of the toxin. If the antitoxin outlived the toxin, the daughter would simply coast free and the whole system would be an expensive bluff.

The known examples are specific enough to check. The F plasmid carries ccdA/ccdB, where CcdB poisons DNA gyrase, jamming it much as a quinolone antibiotic does, and the labile CcdA protein holds it in check. Plasmid R1 carries hok/sok — "host killing" and "suppression of killing" — where the antitoxin is an unstable antisense RNA and the toxin wrecks the membrane potential. The P1 bacteriophage, when it sits quietly as a plasmid, carries phd/doc: prevent host death, and death on curing. The names are the mechanism.

One honest caveat. Toxin-antitoxin pairs are also extremely common on bacterial chromosomes, where there is no plasmid to protect, and what they are doing there is genuinely unsettled — stress response, dormancy and persister formation, and defence against incoming phage have all been argued, with the evidence still moving.

The most tempting misreading is that the toxin somehow stops the plasmid being lost. It does not. Segregation is exactly as sloppy as it was before; daughters keep being born plasmid-free at the same rate. What changes is that those daughters do not go on to found lineages. Selection is not acting on the loss event at all — it is acting on the survivors afterwards. At the population level the plasmid looks stable; at the level of individual divisions nothing has been stabilised.

That distinction explains why real plasmids usually stack several mechanisms rather than relying on this one. Active partition systems physically pull copies to opposite ends of the dividing cell, and multimer resolution systems cut fused plasmid dimers back into monomers so the count stays right. Those genuinely reduce the loss rate. Addiction is the last line — it does not prevent the failure, it only makes the failure sterile.

It is also worth being clear that "addiction" is a metaphor imported into the literature, not a claim about craving. The cell is not dependent on the plasmid in any physiological sense. It is hostage to a poison that happens to be housed there.

• How ccdB was turned into a laboratory tool, killing every cloning host that fails to take up the intended insert.
• Why the same toxin-antitoxin architecture appears on chromosomes and in phage-defence islands, and what that suggests about where these systems came from.
• Whether addiction systems help or hinder efforts to strip antibiotic-resistance plasmids out of clinical populations.

Every term the collection defines is gathered in the glossary.