Every fall, Issaquah Creek in Washington State turns into a maternity ward for thousands of salmon. The fish swim in, spawn, and die, and the water carries away their molecular traces: fragments of DNA, short-lived RNA, chemical marks on the genetic text. From September through November 2024, researchers collected a liter of water every week and ran it through a filter to catch those traces.
The filter then went into a nanopore sequencer, a device the size of a chocolate bar that reads molecules by threading them through a protein pore. But reading alone is not enough: you need to see three things at once. Which species are present, how old the fish are, and what they are sick with. To do that, the RNA was deliberately amplified without any selection, to catch every active gene, while the DNA was left untouched, to preserve the age marks.
The first weeks of the run gave a clear picture. Fresh RNA signaled active life, methylation pointed to adult fish, and pathogens followed their own schedules. The bacterium Aeromonas salmonicida showed up with a one-week lag after the spawning peak, and the oomycete Saprolegnia parasitica multiplied on carcasses two to three weeks later, at one point accounting for more than half of all reads in a sample.
But then the Freshness Gate began to close. RNA vanished within a few days after the fish died, because it is a fragile molecule. Methyl marks lasted longer, but they too were erased, and then the age estimates started to get younger: degraded DNA lost its marks, and the instrument mistook an old fish for a young one. Pathogens, meanwhile, could look alive even when only fragments of them remained.
So the creek water turned out to resemble an old film reel: first the image is sharp, then scratches appear, and finally only noise is left. The work showed for the first time that environmental multi-omics is possible, but only within a narrow time window. Before trusting a methylation-based age estimate, you need to confirm that the sample contains fresh RNA and plenty of long DNA fragments. Otherwise the data lie, and no one will notice the switch.
The practical takeaway for ecologists is simple: take samples in the first weeks of the run, not throughout the season. That saves resources and gives a reliable picture. And Issaquah Creek itself, where the water first told the truth and then began to make things up, remains a reminder that the molecular traces of life have a shelf life.
At the peak of salmon carcass decomposition, the oomycete Saprolegnia parasitica accounted for 51.5% of all reads in the water sample. That is, more than half of the DNA in the stream belonged not to fish but to the fungus eating their remains.
Not published on arXiv. Here are its page, the PDF and the formalization: you can check it yourself, not only read our account.