Pick a leaf. Any leaf — the oak outside, the basil on the windowsill, a fern in a photograph of a rainforest. Healthy, green, unremarkable. Now surface-sterilise it, cut it into squares about two millimetres on a side, and lay the squares on nutrient agar. In a tropical forest, fungus grows out of almost every square. Not one fungus. More than twenty species, from a single leaf, and most of them have no name.
This has been known, in outline, for over a century. It is still not really absorbed. We have spent four issues on fungi as machinery — the kingdom that invented rot, the partnership that built soil, the plumbing under a forest, the body built out of somebody else. This one is about the part of that machinery that is hiding in the most-looked-at object in biology. A leaf is the thing every schoolchild draws. It is also a building with tenants, and we have barely begun the census.
The sentence that opens the field
The standard review of this subject — a Tansley review in New Phytologist, the format the journal reserves for taking stock of a field — opens with a single sentence: “All plants in natural ecosystems appear to be symbiotic with fungal endophytes” Established. Note the hedge in the middle. Appear to be. The authors are not claiming a survey of every plant on Earth; they are reporting that a hundred years of looking has not turned up an exception. The same review puts the history plainly: “Collectively, more than 100 yr of research suggests that most, if not all, plants in natural ecosystems are symbiotic with mycorrhizal fungi and/or fungal endophytes.”
An endophyte, as that review uses the term, is a fungus that “reside[s] entirely within plant tissues and may grow within roots, stems and/or leaves, emerging to sporulate at plant or host-tissue senescence.” Inside, and not making the plant sick. That is the whole definition, and we will come back to how much trouble it causes.
The density is the part that stops people. In lowland forest in central Panama, “100% of mature leaves of diverse trees and shrubs typically contain endophytes,” and “individual leaves may harbor up to one isolate per 2 mm2 of leaf tissue… and frequently contain dozens of species” Established. Scaling up: “Individual plants may harbor hundreds of species, and plant species across their native ranges may be inhabited by thousands of species.”
The gradient from the tropics to the Arctic is, in one respect, extremely tidy. A survey running from Panama to the far north found endophyte incidence falling linearly with latitude — “whereas endophytes were recovered from 95.6–99.5% of 2-mm2 leaf segments examined in lowland, moist tropical forest… only 1% of leaf segments contained endophytes at the arctic site” — a straight line, R² = 0.74. What isn’t tidy is the shape of the diversity. High-latitude leaves hold “relatively few species from many different classes of Ascomycota,” while “tropical endophyte assemblages are dominated by a small number of classes with a very large number of endophytic species” Established. Different shapes of diversity, not just different amounts.
The comparison the same authors reach for is the one worth carrying around. Endophytes from “only a few leaves of a single tropical host species” approach the species diversity of “all plant stems >10 cm diameter at breast height (dbh) on a representative hectare of high-elevation neotropical forest” — their example is Volcan Barva in Costa Rica, at 2,000 m, and the qualifier matters, because high forest is less diverse than lowland forest. Sampling a hundred fungal strains — a hundred fragments, each two millimetres square — recovered diversity “consistent with that of all trees ≥20 cm dbh on the 50-ha Forest Dynamics Plot” on Barro Colorado Island. A hundred confetti-sized pieces of leaf against fifty hectares of trees.
How far from finished is this census? Of 277 species recovered across arctic, boreal, temperate and tropical sites, “138 species (49.8%) were found only once,” and “the species accumulation curve remains non-asymptotic” Established. Half of what they found, they found once. The curve was still climbing when the sampling stopped.
Two ways in
Here is the structural fact that organises everything else. There are two fundamentally different ways for a fungus to get inside a plant, and they produce two entirely different kinds of tenant. (A simplification, and we should say so on the way in: the standard review actually sorts endophytes into four classes, and its Class 2 uses both routes. The clean binary below is the two ends of a range, not a partition.)
Inherited. A small group of endophytes lives in grasses and travels in the seed. The review calls these Class 1: “a small number of phylogenetically related clavicipitaceous species… limited to some cool- and warm-season grasses,” forming “systemic intercellular infections.” Their transmission “is primarily vertical, with maternal plants passing fungi on to offspring via seed infections,” and “colonized plants usually harbor one dominant fungal isolate/genotype” Established. One fungus, throughout the plant, handed down the maternal line like a piece of furniture.
Caught. The endophytes of tree leaves do the opposite. They are absent when the leaf unfurls and arrive from outside. Grow cacao seedlings under clean conditions and endophytes appear “in <1% of tissue segments of mature leaves”; the seeds carry none. Meanwhile, in lowland Panama, “>10,000 fungal propagules are deposited as air spora on leaf surfaces each day”, and “leaves accumulate numerous infections shortly after emergence by means of epiphytic germination of fungal propagules, followed by cuticular penetration or entry through stomates” Established. The review reports, citing a companion field study, that endophytes were present in “> 80% of Theobroma cacao leaves within 2 wk of leaf emergence.”
So: one lineage of plants inherits its fungi, and another catches them, the way you catch a cold, within a fortnight of opening a new leaf. The consequences run all the way down. Inherited endophytes are low-diversity and systemic. Caught endophytes are hyperdiverse and, as the same review puts it, form “highly localized infections.” And — this matters for what follows — the inherited ones are the ones we noticed first, because one of them was poisoning cattle.
The grass that had to be caught out
Tall fescue is, in a phrase from the agronomist who spent a career on it, “the most important cultivated pasture grass in the USA.” For decades, cattle grazing it in the American South did badly in ways nobody could pin down: rough coats they wouldn’t shed, poor weight gain, low conception rates, and in bad cases dry gangrene of the tail tip, ear tips and hooves. It has an unlovely name — fescue foot.
In 1977, Charles Bacon and colleagues found the suspect, and the reason it had gone unnoticed is the whole point of this issue. In their words: “The wide-spread occurrence of E. typhina as an endophyte in tall fescue has been unnoticed previously because the infections are symptomless. Consequently, E. typhina has not been investigated as a possible cause of the fescue toxicity syndrome” Established. The fungus had been invisible because it was doing its job well. The correlation was stark: “The frequency of E. typhina in toxic fescue samples was 100% regardless of the type of grass, but in nontoxic samples it varied from 0 to 50%.”
Worth pausing on how carefully they wrote it up. They did not announce a cause. They wrote that the relationship “suggests the possibility of involvement of E. typhina in the fescue toxicity syndrome,” and noted what further work would be needed. It took the field years to close that gap; the modern statement is direct. Ergovaline “has been recognized as the most abundant and the most toxic ergopeptide alkaloid produced in E. coenophiala-infected tall fescue, and is responsible for fescue toxicosis in livestock” — the alkaloid is now generally held responsible, and the reviews that work on the physiology agree Established.
The mechanism is a case of chemical mistaken identity. The alkaloids work “largely due to the structural similarities of the ergoline ring structure to the biogenic amines, serotonin, dopamine, norepinephrine, and epinephrine,” which lets them “bind biogenic amine receptors and to elicit such effects as decreased serum prolactin and vasoconstriction” Established. Squeeze the blood vessels and the extremities go — “tissue necrosis or dry gangrene of the ear tips and tail… and, in extreme cases, loss of affected hooves.” Suppress prolactin and the animal doesn’t shed its winter coat, so it “will maintain a shaggy coat, even in the heat of summer, which further exacerbates heat stress.” A fungus inside a blade of grass, reaching a cow’s hair follicles through her pituitary. (The same chemical family has a much older and darker human history, under a different fungus and a different name. That is its own issue, and we’re not spending it here.)
The fungus does this for reasons that have nothing to do with cattle. The genus makes four alkaloid classes, and they divide the work by target: ergot alkaloids and indole-diterpenes “are active in vertebrate systems, and lolines and peramine… are more specific against invertebrates” Established. Insect deterrents and mammal deterrents, on the same payroll. The cattle are collateral.
The part that makes it a bargain
Here is where a simple story about a toxic pasture becomes something better. The obvious fix — breed the fungus out — was tried, and it failed in an instructive way. Endophyte-free tall fescue “could not persist through drought and heavy grazing,” and “selection for more persistent endophyte-free varieties was attempted but was unsuccessful.” In one moisture-stress trial, “75% of the E- plants died, while all the E+ plants survived” Established. Take out the fungus and you lose the pasture. The toxin and the toughness came in the same package.
What worked instead was swapping the strain. Researchers found naturally occurring endophyte strains that make the insect-deterring alkaloids but little or no ergot alkaloid, and inserted them into good fescue cultivars — literally, by making “a small longitudinal slit in an E- seedling just below the apical meristem and inserting fungal mycelia.” The 2002 trial that established the approach reported that the best combination had stand survival “equivalent… to the E+ checks,” with lamb weight gains “approximately 57% greater than gains on E+ forage” Established. These are sold as “novel” or “friendly” endophyte cultivars, and extension services are careful to distinguish them from endophyte-free seed, which is a different and worse thing.
And because the fungus only rides in the seed, the whole system has a strange fragility. “Spreading through seed is the only known means of transmitting the fungus to the next generation. An uninfected tall fescue plant cannot ‘catch’ the fungus from an infected plant” Established. Which means the fungus can be lost simply by leaving the seed on a shelf: stored under ambient conditions, endophyte viability falls to zero in roughly 18 to 24 months, and the seed itself outlives it — “the endophyte would die, but the seed remained viable.” An unbroken maternal line, ended by a warm season in a warehouse.
A note on the numbers around this grass, which are worse than the biology. Losses to US beef production are cited everywhere; the trail is a mess. A 1993 assessment is quoted by later authors as “well over $600 million annually,” and by others as “almost 600 million dollars” — we could not open that 1993 paper, so we are quoting the people who cite it, and one widely-circulated version puts the underlying data in 1990 rather than 1993. Downstream, that becomes “likely exceed $1 billion” — an author’s explicit guess, not a calculation — and then “over $2 billion,” attributed to a 2015 paper whose own published figure caption describes a cow–calf-sector calculation only. Acreage is quoted as 35, 37 and 40 million acres by different sources, the earliest traceable citation being a book chapter from 1979. And the “more than 90% of pastures are infected” line is routinely hung on a 1987 survey that reports a mean infection of 58% from voluntarily submitted samples. One survey did measure it properly: in Georgia in 2022, across 21 counties, “the mean infection level for the whole study was 94% with a median of 100%” Established — that stamp being on the Georgia measurement alone, in 56 samples, in one state, and on nothing else in this paragraph. The biology here is solid. The economics is a game of telephone, and we would rather show you the wire than repeat the last thing down it.
One more thing to get right, because this brand keeps running into it: the fungus has changed its name three times. Called Epichloë typhina in 1977, it was renamed Acremonium coenophialum in 1982, moved to Neotyphodium coenophialum in 1996, and returned to Epichloë coenophiala in 2014 — after the 2011 botanical congress ratified “one fungus = one name” and stopped giving a fungus separate names for its sexual and asexual states Established. The grass is worse. As we checked these records in September 2026, it was placed in Schedonorus by USDA PLANTS and ITIS, and in Lolium by Kew’s Plants of the World Online and GBIF — not a settled succession but a live disagreement between major authorities about the name of the most important pasture grass in America. That is our own reading of four database records that contradict each other; we found no paper describing the split as a dispute, and databases move, so treat it as a snapshot rather than a permanent state of affairs.
The experiment on the other side of the split
If inherited endophytes earn their keep, what about the caught ones — the hyperdiverse crowd in a tree leaf, arriving by air, colonising a few cells and stopping? For a long time the assumption was that they did little. As one paper phrased the prior view: “the diverse, horizontally transmitted endophytes of woody angiosperms are thought to contribute little to host defense.”
In 2003 a team tested it directly on cacao. They raised endophyte-free seedlings, inoculated some leaves and not others on the same plant, then challenged both with a Phytophthora strain isolated from diseased cacao in Panama — a water mould, an oomycete, not a fungus at all. The result: “leaves lacking endophytes experienced leaf death and abscission 2.8 times more frequently than did leaves inoculated with endophytes.” On leaves that survived, necrosis covered “12.01 ± 1.36% of leaf area” without endophytes against “6.20 ± 1.28%” with them. Across the whole plant, photosynthetic surface fell 32.3% in the endophyte-free treatment and 14.1% in the treated one Established. Fungi the leaf caught out of the air were doing something a leaf would pay for.
Three honest qualifications, two of them from the authors themselves. They flag that “by introducing a large inoculum volume of Phytophthora sp. through cuticular wounds, we provided advantageous conditions for infection” — this was a hard challenge, not a natural one. They flag their own open question about “the potential for endophytes to confer hidden costs on their hosts.” And a third we’ll add: the headline mortality result rests on a paired t test with a single degree of freedom, which is thin. The necrosis result, based on 135 degrees of freedom, is the sturdier number.
The follow-up work is the interesting part, and it moved the story. A 2014 paper from the same group reports that the benefit is now attributed in part to one dominant species priming the plant’s own defences rather than to fungal diversity or direct chemical warfare — “Thus, far, no evidence suggests chemical inhibition of Phytophthora spp. by C. tropicale” [sic, comma and all] — and that the hidden costs the 2003 team flagged as an open question now at least have names. Those authors put it carefully, and we’ll keep their hedge: the benefits “appear to be offset by potential costs including reduced photosynthesis, altered host nitrogen metabolism, and endophyte heterotrophy of host tissues” Established. The emphasis on potential is ours, and it is doing real work — these are costs identified, not costs settled. The bodyguard looks real. It also appears to eat.
The one that shouldn’t work
And then there is the case that sounds made up.
In the geothermal soils of Yellowstone and Lassen Volcanic National Parks grows a panic grass, Dichanthelium lanuginosum, in ground that runs from about 20°C to 50°C over the year. In 2002 a team heated the root zones under laboratory conditions and watched: “nonsymbiotic plants (45/45) became shriveled and chlorotic at 50°C… In contrast, symbiotic plants (45/45) tolerated constant 50°C soil temperature for 3 days and intermittent soil temperatures as high as 65°C for 10 days. All nonsymbiotic plants (45/45) died during the 65°C heat regime, whereas symbiotic plants (45/45) survived” Established. The difference was a fungus living inside the grass, later named Curvularia protuberata. Neither partner manages it alone: “When grown separately, neither the fungus nor the plant alone is able to grow at temperatures above 38°C, but symbiotically, they are able to tolerate elevated temperatures.”
Which is already a good story. In 2007 it got a third character. The fungus carries a virus — two segments of double-stranded RNA in spherical particles about 27 nm across, named Curvularia thermal tolerance virus. Cure the fungus of the virus and the arrangement collapses: plants inoculated with the virus-free fungus “became shriveled and chlorotic and died,” exactly like plants with no fungus at all. Put the virus back and the heat tolerance comes back. The authors’ own summary: “Fungal isolates cured of the virus are unable to confer heat tolerance, but heat tolerance is restored after the virus is reintroduced” Preliminary. A plant that survives 65°C because of a fungus that can do it because of a virus.
We’ve stamped that Preliminary rather than Established, and here is exactly why, because the reasons are specific rather than a general shrug. The heat tolerance itself is solid — big, replicated, 45 plants out of 45. The virus requirement is not equally solid. The reintroduction experiment worked by fusing two fungal strains and screening the offspring: “Thirty-five hygromycin-resistant isolates obtained in this way were screened for their dsRNA profiles, but only one was found to have acquired the virus.” One isolate carries the entire complementation arm. The paper’s proposed mechanism failed its own test in the same paper — “there was no difference in the ROS response to heat between plants inoculated with the virus-free and the CThTV-infected isolates.” And we could find no independent laboratory that has repeated any of it: no failed replication, no rebuttal, no retraction, but also no outside confirmation. And the famous “neither survives above 38°C” line, which the 2007 paper hangs on the 2002 one, has no direct counterpart in that earlier paper: the only comparable figure there is 40°C, and it refers to the fungus alone in pure culture, not to both partners. Not a contradiction — two different measurements — but the 38°C number that gets repeated everywhere is the 2007 restatement, not something the 2002 experiment measured.
None of that means it’s wrong. It means one lab, one strain pair, one successful isolate, and an unexplained mechanism. That is a wonderful result at an early stage, and calling it anything else would be doing the reader a disservice.
Where we get honest
First: are these fungi bodyguards at all? The tidy version of the grass story — the fungus makes toxins, the toxins deter herbivores, so the fungus is a defensive mutualist — has a serious and long-running critic. Stanley Faeth’s 2002 argument is that the whole framework was built on the wrong organisms: “Most of the evidence for the defensive mutualism concept is derived from studies of agronomic grass cultivars, which may be atypical of many endophyte-host interactions. I argue that endophytes in native plants, even asexual, seed-borne ones, rarely act as defensive mutualists” Contested. In wild grasses, he notes, infected plants “vary enormously in alkaloid levels, from none to levels that may affect herbivores,” and infection frequencies form mosaics rather than the near-100% of a managed pasture. He and a colleague went further in 2003, in a paper whose title does the arguing: Mutualistic Asexual Endophytes in a Native Grass Are Usually Parasitic. Infection “generally decreased host growth… decreased reproduction… and the seeds produced by infected plants had lower germination success.”
Keith Clay, who named the defensive mutualism in 1988, has the strongest empirical answer: over 54 months, endophyte infection frequency rose significantly more where herbivores were present (30%) than where they were fenced and sprayed out (12%) — herbivory selecting for the fungus in real time. But notice the grass in that experiment is tall fescue, an introduced agronomic species: precisely the system Faeth says is unrepresentative. Two decades on, the honest description is not that one camp beat the other but that they have been studying different plants and publishing in parallel. We looked for a formal reply from Clay to Faeth’s 2002 critique and could not find one; no comment-and-reply pair appears to exist.
The nearest thing to a referee is a 2010 meta-analysis of 99 studies, and its verdict is deflating to both sides: “Consistent with the defensive mutualism hypothesis, the meta-analysis indicates that endophytes slightly increase grass resistance to herbivores,” with the effect clearest in exactly the vertically-transmitted grass systems and weakest in tree endophytes — but “current literature is insufficient to capture the breadth of variability inherent in the wild grass-endophyte populations and communities, and the general importance of defensive mutualism remains to be solved in future studies.” Our emphasis on slightly. One awkward disclosure while we’re here, because it undercuts our own use of the word referee: two of that meta-analysis’s three authors co-wrote, with Faeth himself, the 1998 review that introduced the “antagonistic to mutualistic continuum” framing now widely treated as the neutral middle ground. It is a good framing, and the meta-analysis is a careful piece of work. Neither is a disinterested bench.
Second, and larger: what does the word mean? We have been using “endophyte” for this entire issue as though it named a kind of organism. It probably doesn’t. A major review of microbial endophytes — written mostly about bacteria, in a bacteriology journal, and arguing for a term that covers all microorganisms — says so in its own voice: “we question the currently applied definition of endophytes and claim that the term ‘endophyte’ should refer to habitat only, not function” Philosophy. A mycological review makes the same move from the fungal side: the term “refers only to fungi at the moment of detection without regard for the future status of the interaction.”
Read that carefully. An endophyte is not a lifestyle. It is a fungus you happened to find inside a symptomless plant on the day you looked. The category is defined by two negatives — inside, and not visibly causing disease — and both are observer-relative. Whether something counts as “not causing disease” depends on whether anyone ran the test: as those authors note, pathogenicity assays “in most cases… are not performed, or they are performed with only one plant species, although pathogenicity might occur with a different plant genotype or under different conditions.” The boundary of the category tracks how hard people looked.
We’ve stamped that Philosophy deliberately, and it’s the second time this series has used the rung, so the reasoning should be on the record. What is inside a leaf is empirical, and more sequencing will settle it. Whether the word “endophyte” should name a place or a job is a decision about how to carve up a continuum — and both sides are proposing a stipulation, not reporting a measurement. As with the lichen definition fight last issue: definitional at the surface, empirically gated underneath.
The best-known attempt to give the term some functional content is worth handling carefully, because it is routinely quoted as an established mechanism and it is not one. The proposal is that endophytism is a standoff — “We hypothesize that as long as fungal virulence and the host defence reaction are balanced, disease does not develop” Preliminary — so that the same fungus turns pathogenic if the plant weakens, and saprotrophic if the plant dies. The observations behind it are real: endophytes and pathogens “both possess many of the same virulence factors,” and the host “can react with the same defence reactions as to a pathogen.” But every load-bearing sentence in that literature opens with we hypothesize, the key figure is captioned “Hypothesis,” and the authors state the gap outright: “it is unclear how this balance is regulated.” The secondary literature has a bad habit of quoting the sentence with its opening clause removed, which is exactly how a hypothesis gets laundered into a fact. We’d also note the frame is close to unfalsifiable as usually deployed — quiet fungus, balance holding; sick plant, balance tipped; dead plant, balance tipped the other way. Every outcome confirms it. That is a sign of a useful way of organising cases, not a tested mechanism.
Third: do the tenants become the wreckers? There is a lovely idea here, and it connects straight back to the first issue in this series. If the fungi that decompose a leaf are already living inside it, then rot doesn’t arrive — it wakes up. There is genuine evidence: endophyte and saprotroph isolates from the same host have been found phylogenetically interleaved and producing the same degradative enzymes, leading those authors to conclude — in a follow-up three years later, not the original phylogenetic study — that there is “further compelling evidence that endophytes can switch lifestyle to saprobes,” and a 2017 study found previously-endophytic fungi “by far the most active fungi in 1-year-old litter” Contested.
And there is genuine evidence the other way, which we went looking for on purpose. A cross-biome study by two of the leading endophyte researchers states flatly that “the prevalence of saprotrophic life phases among endophytes and the dynamics of such fungi on the endophyte-to-saprotroph continuum are not yet clear,” and found “consistent differences between fungal assemblages in living vs. non-living leaves, reflecting incursion by fungi from the leaf exterior after leaf death.” Their own data did not find that living-leaf fungi differ consistently in cellulose-degrading ability from dead-leaf ones. The authors of the switching studies hedge too — “possibly exist,” “some endophytic species.” The defensible claim is narrow: some lineages demonstrably persist into litter and degrade it. The sweeping version — that endophytism as such is a dormant decomposition phase — is not supported.
And here is why the word problem is not just semantics. It is quietly wrecking one of the headline numbers in mycology. If you count the same fungus once as an endophyte in a living leaf and again as a saprotroph in the litter beneath it, you have inflated global fungal diversity with a bookkeeping artefact — and the researchers above name the mechanism, noting that blurring these categories creates “a challenge for estimating trophic modes and the scale of fungal diversity.” You can watch it happen. The 2000 argument that excluding endophytes “will markedly underestimate fungal species diversity” fed a ratio-based extrapolation implying over 1.5 million fungal species; a 2023 reassessment ran the same approach through its own correction and got half a million out of it — not a global total, but a measure of how much the method moves under pressure. Meanwhile a 2022 metabarcoding study put global richness at 6.28 million, and that same 2023 reassessment deflated the figure to 1.66 million by arguing the sequence-clustering threshold inflates taxa by roughly 3.8 to 1. The reassessment’s own answer is “2–3 million fungal species with a ‘best estimate’ at 2.5 million” Contested — and a 2024 review declined to issue a number at all, concluding that a more accurate account “was not possible based on present evidence.” For scale: the Catalogue of Life currently accepts 157,091 fungal species. Take the three live global candidates — 1.66, 2.5 and 6.28 million — and between about 90% and 97.5% of the kingdom is unnamed, the width of that range being entirely a function of whose method you trust. A real share of what’s missing is inside the leaves outside your window.
Which leaves the question this issue keeps backing into and won’t answer today. If every leaf on a tree is a consortium — if the plant’s defences, its drought tolerance, its heat tolerance and possibly its decomposition are all partly outsourced to organisms it caught out of the air — then where exactly does the tree stop? We reached the same edge from the lichen side last issue and declined it then too. It has a whole arc coming in this plan, and it deserves the room. For now: you have never seen a leaf. You have seen a leaf and its tenants, and nobody has finished the census.