The Plant That Counts To Five

About eight hundred species, and carnivory was invented at least twelve separate times. A carnivorous plant is defined by doing four things: attracting prey, capturing it, digesting it, and absorbing the nutrients. Roughly eight hundred species qualify, and they are scattered across the flowering plants in a way no single origin could produce. Caryophyllales gives us Drosera the sundews, Dionaea the Venus flytrap, Nepenthes the pitcher plants, and Drosophyllum. Ericales gives us Sarracenia and Darlingtonia. Lamiales gives us Pinguicula, Utricularia the bladderworts, and Genlisea. Poales gives us Brocchinia. Oxalidales gives us Cephalotus, on its own. These groups are not close relatives, and the traps they built are the same four designs: pitfall, snap, adhesive and suction.

Darwin gave them a whole volume. Insectivorous Plants, John Murray, 1875. He fed sundews fragments of meat and egg white and hair and glass, timed how long the tentacles took to bend, and worked out that the plant could distinguish nourishment from debris. Of the Venus flytrap he wrote that it was one of the most wonderful plants in the world. It is worth noting a small thing about how our own file cites this: the entry carries no digital identifier at all, which is exactly right for an 1875 book, and it does not reach for the identifier of some later reprint to fill the space. We will come back to why that is worth remarking on.

And the trap counts. Closure is set off by trigger hairs, typically three on each lobe. One touch does nothing. Two touches within about twenty seconds produce the electrical signals that fire the snap, which is a straightforward filter against raindrops and falling debris. But a 2016 paper in Current Biology showed the counting goes further than that. Two touches close the trap. Three activate the sealing response. Five or more start the secretion of digestive enzymes. The plant is tallying prey-induced action potentials and changing its behaviour at each threshold, so that it only pays for digestion once something has struggled enough to be worth digesting. Our file's phrase for this is that Darwin's plant counts to five. One correction to the record: our file's bibliography gives the first author's given name as Jörg. It is Jennifer.

And the whole species lives in one small place. Dionaea muscipula occurs naturally within roughly a hundred and twenty kilometres of Wilmington, North Carolina, and nowhere else. The most recognisable plant in the world, the one every child can draw from memory, has a native range you could drive across in an afternoon.

Bladderworts fire in about half a millisecond. That is roughly two hundred times faster than the flytrap, the fastest known movement in the plant kingdom, and among the fastest movements in any organism at all. Utricularia grows tiny underwater bladders, each with a hinged door. The bladder pumps water out through specialised glands until its walls are held inward under a pressure difference of something like 1.5 megapascals. Trigger hairs on the door are disturbed, the door buckles open, and the wall springs back to its resting shape, drawing prey and water in before the animal can react. Philippe Marmottant, Olivier Vincent and colleagues filmed the whole sequence at fifteen thousand frames per second and published the analysis in 2011. There are about two hundred and thirty Utricularia species, which makes it the largest genus of carnivorous plants and, by a wide margin, the least famous.

Proteases, phosphatases, esterases, ribonucleases and chitinases. The chitinases matter especially, because an insect's exoskeleton is chitin and something has to open it. A 2012 proteomics study identified more than thirty proteins in the fluid, including aspartyl proteases, glucanases and peroxidases. One correction to our own file here, and it is a real one: the body of the file says that study characterised the fluid of Nepenthes pitcher plants. It did not. The paper is titled after the digestive fluid of the Venus flytrap, and the flytrap is what it examined. Our file's own bibliography entry has the title right, which means the error is in the body and the bibliography is the half that is correct. The nepenthesins that the same passage attributes to that paper are indeed Nepenthes enzymes, from a different body of work altogether. Two literatures ran together in one sentence.

And the prey really does become the plant. Feed a carnivorous plant insects labelled with nitrogen-15 and then measure where that isotope ends up, and you find that prey-derived nitrogen accounts for somewhere between twenty-five and seventy-five percent of the total nitrogen in the leaves, depending on the species and how much prey is available. Absorption happens through specialised glandular hairs on the trap surface. This is not a supplement at the margins. In some species, most of the plant's nitrogen walked in on its own legs.

The signalling is electrical, and the resemblance to animal nerves is not a metaphor. Venus flytraps and sundews generate action potentials that look remarkably like the ones in animal neurons, propagated along cell membranes by voltage-gated ion channels. Rainer Hedrich and Erwin Neher set out the machinery in Trends in Plant Science in 2018: mechanosensitive channels at the trigger hairs convert a physical bend into a voltage change, anion channels carry the signal, and glutamate receptor-like channels are involved in propagating it. Then there is the detail that makes the whole thing feel less like an invention and more like a repurposing: the digestive enzymes are switched on by jasmonic acid signalling, which is the same hormone pathway plants everywhere use to defend themselves against being eaten. The plant did not build a new system for eating animals. It turned the anti-herbivore alarm around. Alexander Volkov's name for the trap is a green muscle, an electrically controlled actuator with no muscle in it anywhere.

A trap is an expensive leaf, so it only pays in poor soil. Aaron Ellison and Nicholas Gotelli formalised carnivory as a cost-benefit calculation, and it is the framework most of the field now uses. A trap is a poor photosynthesiser compared with an ordinary leaf: the surface is given over to glands and glue and hinges rather than to catching light. So a plant will only build one where the nutrient return exceeds the photosynthetic cost, which happens precisely where soil nutrients are scarce and light is not. That predicts the habitats these plants actually occupy: bogs, fens, sandy wetlands, seepage slopes, exposed mountaintops. And it predicts an experiment that works. Fertilise a carnivorous plant with nitrogen and phosphorus and many species respond by building fewer traps and more conventional leaves. Given a cheaper source of nitrogen, the plant stops paying for the expensive one.

And the objection is in our own file, stated fairly. Méndez and Karlsson argued in 2005 that the cost-benefit framework oversimplifies, because it treats traps as photosynthetically wasteful when in genera like Drosera and Pinguicula the traps are the photosynthetic leaves. There is no separate ordinary foliage to compare them against, so the cost side of the ledger is harder to price than the model assumes. Our file records this objection in its own counter-arguments section rather than leaving it out, which is the behaviour you want from a source. We should say plainly that we did not chase down the Méndez and Karlsson paper ourselves; it has no bibliography entry in the file, and we did not verify it.

They may work as a gauge for the health of a wetland. The argument runs like this: because carnivorous plants occupy such narrow conditions, wet, acid and nutrient-poor, they should be unusually sensitive to the things that change those conditions, which are climate shift, atmospheric nitrogen deposition and drainage. That would make them useful early indicators of wetland degradation. The habitat pressure is not in doubt: Dionaea muscipula is listed as Vulnerable on the IUCN Red List, and its native habitat is estimated to have shrunk by ninety-three percent since European settlement through fire suppression, drainage and building. In 2014 North Carolina made poaching wild Venus flytraps a felony. What has not been done at scale is the monitoring itself, so the bioindicator idea remains a proposal with good reasoning behind it rather than a demonstrated method.

No plant has ever eaten anybody, and we know exactly where the story started. The man-eating tree of Madagascar was invented in 1874 by Edmund Spencer, writing in the New York World. It was fiction presented as a traveller's report, and it has been recirculating ever since, eventually producing Audrey II. The real ceiling is much lower and much less dramatic. The largest traps in the world belong to Nepenthes rajah, whose pitchers reach about thirty-five centimetres, and they do occasionally catch small vertebrates: frogs, lizards, the odd shrew or mouse. That is the maximum. Nothing about the structure or the energetics of a plant supports anything beyond it. There is no mechanism by which a plant could hold and process an animal of any size, and there is no organism that has ever tried.

We resolved every identifier in this file live, and six of the eleven are exact. Forterre on the snap, the Current Biology paper on the counting, Vincent on the suction traps, Ellison on nutrient stoichiometry, Chase on murderous plants, and Hedrich and Neher on how the flytrap works all resolve to precisely the works our file names, with the right journal, volume, issue and pages. That is a real majority, and it is worth saying before the rest of this section, because the rest of this section is about the other five.

Every failure in this file has the same shape: the right journal, the wrong article. The Phytochemistry pointer lands on a paper about geraniol synthase in Madagascar periwinkle, at pages 36 to 43, immediately before the 44 to 58 our file claims. The Trends in Ecology and Evolution pointer lands on a paper about recombination in sex chromosomes, in the same volume, the same issue, with an overlapping page range. The Journal of Experimental Botany pointer lands on a paper about the construction costs of leaves. The Frontiers in Microbiology pointer lands on a paper about yeast. And the Nature Ecology and Evolution pointer lands nowhere at all.

That last one is off by a single character. Our file gives the Cephalotus genome paper the identifier 10.1038/s41559-017-0059. That is a flat 404; nothing is registered there. The real identifier is 10.1038/s41559-016-0059, and it resolves to exactly the paper our file names, by Kenji Fukushima and colleagues. One digit. This is the second single-character identifier error we have found this month, after a reference in our hadal zone file where the same slip appeared in both the identifier and the volume number. And in both cases the wrong character killed the pointer outright rather than redirecting it, which is by some distance the better failure: a dead link tells you something is wrong, and a working link to the wrong paper does not.

One entry is a chimera, and the file's own body knows better. The single Ellison and Gotelli entry in the bibliography carries the title of their 2001 paper in Trends in Ecology and Evolution, the year of their 2009 paper in the Journal of Experimental Botany, and a volume and page range belonging to neither. Both papers are real; we found and verified both. And here is the part worth pausing on: the file's own prose cites them correctly and separately, quoting Ellison and Gotelli 2001 for the nitrogen isotope figure and Ellison and Gotelli 2009 for the energetics argument. The body of the document knows there are two papers. The bibliography collapsed them into one.

Four of the five were recovered, and the fifth we will not pronounce on. The Fukushima genome paper, the digestive fluid proteomics paper, both Ellison and Gotelli papers and the bladderwort paper all exist and were found; the correct identifiers for all of them are in the source list at the foot of this article. The proteomics recovery is the largest single correction, because it moves the paper out of Phytochemistry entirely and into Molecular and Cellular Proteomics, changes the volume and pages, and gives the first author's given name as Waltraud X. where our file says Wolf B. The one we could not resolve is a paper attributed to Leonora Bittleston in Frontiers in Microbiology. A title search returns nothing matching, and an author search across five years returns her real output on exactly this subject but no paper of that title in that journal. We record it as not found. That is not the same statement as does not exist, and we are not making the stronger one.

And this file's failure profile is the mirror image of the others. Over the past month we have resolved the identifiers in eight research files one at a time. In the files whose sources are mostly books, every single bad identifier turned out to point at a published review of the book rather than the book: five of five in the Nizari Ismaili file, five of five in the Book of Enoch file, three of three among the Olmec volumes. In the files whose sources are mostly journal articles, not one bad identifier is a review; every one lands on a neighbouring paper. The hadal zone file, the Hittite file, and now this one at five out of five. The explanation is mechanical rather than mysterious. A book has no neighbouring article, so a pointer that drifts from a book has nowhere to land except on something else bearing its title, and the only thing that bears a book's title is a review of it. A journal article's nearest neighbour is the paper printed beside it in the issue. Whatever process produces these errors, the shape of the wreckage is decided by the shelf.

This file is also the best-behaved one we have audited, in three specific ways. Its bibliography is properly fielded: no author list is cut in half at a middle initial, where three other files this month carried eleven entries between them shattered that way. Its cross-references pass the label audit: no link in it is labelled with the name of one document while opening another, which is a defect we measured at up to sixteen percent across the library as a whole. And its Darwin entry carries no identifier, where another file put a 1988 reprint's identifier on an 1875 book. One detail may explain all three at once, or may not: this file was last updated on the twenty-fifth of June 2025, which makes it the oldest document in the whole audit. Whatever produced the split author names and the borrowed reprint identifiers in the 2026 files is not present in this one.