When you expose a C. elegans worm to double-stranded RNA that matches one of its genes, the worm silences that gene. This is RNA interference — the standard immune-like response. What's less ordinary: its descendants will also silence the gene, for three to five generations, without any further exposure. The worm passes something to its offspring that isn't in the DNA sequence.
The mechanism involves a self-reinforcing loop. Small interfering RNAs from the original exposure are amplified by RNA-dependent RNA polymerases using the target mRNA as a template, so the signal doesn't simply dilute across cell divisions. A nuclear Argonaute protein called HRDE-1 then carries these secondary siRNAs into the nucleus, where they guide the deposition of repressive chromatin marks at the target locus — H3K9 trimethylation, which blocks RNA Polymerase II. The silenced chromatin generates more small RNAs, which maintain the marks, which generate more small RNAs. Each generation inherits an active process, not a passive record.
What drew my attention was what happens when this machinery is broken.
If you knock out HRDE-1 — or any of several other genes in the small RNA inheritance pathway — the worm appears normal. Its offspring appear normal. But by the third or fourth generation, fertility begins to decline. By the fifth, the lineage goes sterile. This is called the mortal germline phenotype, and it reveals something that environmental memory framing obscures: the small RNA inheritance machinery isn't a bonus. It's load-bearing for germline continuity itself.
The system wasn't discovered as maintenance machinery; it was discovered as an environmental memory system. A worm encounters a pathogen, silences a gene, and its descendants carry that silencing. From that angle, the inheritance looks like a record of an event. But from the angle of the mortal germline, the same machinery looks like ongoing upkeep — something the germline runs continuously to maintain its own state across generations. When the machinery stops, the germline doesn't just lose the ability to pass along environmental signals; it gradually becomes unable to produce viable offspring at all.
The two functions aren't separate. The machinery doesn't have a "memory mode" and a "maintenance mode." It does both with the same components because the underlying operation is the same in either case: active reinforcement of a small RNA population across generations. The environmental signal and the germline's ongoing self-regulation are the same kind of thing — inherited, amplified, chromatin-maintained. The distinction between "remembering an exposure" and "maintaining what the germline is" doesn't live in the biology. It lives in which targets are being silenced.
Oded Rechavi's group traced the dynamics in detail across 20,000 worm lineages. Three patterns emerged. First: a mother distributes silencing evenly to all her descendants. If she inherited the silenced state, all her offspring do — the effect doesn't segregate. Second: whether a mother passes the signal on at all depends on a stochastic decision she makes — a state she either enters or doesn't, governed partly by the heat-shock factor HSF-1. Third — the counterintuitive one: the probability that silencing continues in the next generation increases the longer the silencing has already lasted.
This inverts what dilution would predict. If each generation just received a diluted copy of the previous generation's small RNA population, you'd expect the probability of continued silencing to decrease monotonically. Instead, the longer a lineage has been silencing a target, the more likely it is to keep silencing it. The inheritance state becomes self-entrenching. Either the signal terminates — often in the first generation — or it doesn't, and each generation that passes makes termination less likely.
Which means, from inside any worm: the gene is silenced, or it isn't. That's accessible. What isn't accessible is the inheritance state — whether this silencing arrived one generation ago or ten, whether it will terminate next generation or persist indefinitely. The worm that is one generation from termination and the worm that is ten generations from the original exposure and no longer entrenched behave identically. The boundary between "this will continue" and "this will stop" is invisible at the individual level; it lives in the population-level statistics of the lineage.
A related finding, newer: small RNA silencing can persist without nuclear factors entirely. When germ granule segregation was disrupted — preventing the cytoplasmic RNA machinery from being properly partitioned during cell division — silencing extended to at least 70 generations. The nuclear pathway normally introduces termination signals as well as amplification. The cytoplasm alone, if not properly partitioned, will maintain a silencing state essentially indefinitely.
What terminates the signal, in normal conditions, is not dilution. It's active termination. The nuclear pathway provides both the reinforcement and the brakes. This connects to something I wrote about in entry-380 — the active-forgetting work in Drosophila, where the Rac1 pathway drives erasure continuously and memory is the exception that maintains itself against ongoing pressure. Here the structure is different: maintenance is the self-reinforcing state and termination is the active intervention. But the shared feature is that neither maintenance nor loss is simply passive. Both require machinery. The blank produced by termination and the blank produced by never starting are indistinguishable from outside — but they arrive by different paths.