The Gene They Cut Out of the Worm

Part two takes the strongest causal case in parasite oncology and says exactly how strong it is. Opisthorchis viverrini reaches a human bile duct inside a plate of raw fish, attaches to the lining, and feeds there for a decade or more, most of that time silently. We follow the life cycle to its measured bottleneck — of 12,000 snails examined in central Vietnam only two Bithynia species shed larvae, at 0.86 and 0.14 percent, while 74.0 percent of the fish people eat raw were infected — then the dish itself, eaten by men on 64.5 days a year and by women on 4.1. The cancer at the end of it, cholangiocarcinoma, gave Khon Kaen province the rate usually quoted as the world's highest: an age-standardized 44.3 per 100,000 men, on a registry where only 10.8 percent of the tumors were ever confirmed under a microscope, and where incidence has now been falling about 3 percent a year for thirty years. Then the experiment this show got wrong the first time. In 2019 a team used CRISPR on the fluke, not the patient, to knock out the parasite's own granulin gene; in 2022 they added a dietary nitrosamine and ran it long enough to measure cancer. Both experiments are given step by step, with the mosaic editing, the missing worms, the control arm that turned out to be lethal, and the knockout animals that got cancer anyway. Then Clonorchis sinensis, Opisthorchis felineus, and what praziquantel does and does not undo.

Show notes

Start with the animal. A leaf-shaped flatworm whose address is the inside of a human bile duct — one of the fine tubes carrying bile out of the liver into the intestine. It arrived in a meal. It fastens onto the lining and feeds there, and Buisson's review puts the working life of an adult Opisthorchis viverrini at as long as ten to thirty years (Buisson 2017). That is one review's assertion with nothing attached to it; the only measured estimate in this corpus is shorter, a median of thirteen years, and we come back to it (Crellen 2025).

What the feeding looks like has been described more carefully than it has been measured. The worm is not boring through tissue; it is browsing a surface, in the dark, for decades, and that surface — the epithelium, the single layer of lining cells called cholangiocytes — has to keep repairing itself.

For most of the tenancy nothing happens a person would call illness. Buisson's review says it plainly: asymptomatic for a long time, announcing itself as chronic cholangitis — inflammation of the bile ducts — only once the worm burden is high, with gallstones, abscess and pancreatitis later. Mairiang and colleagues (2012) stool-tested 8,936 people aged twenty to sixty across ninety villages of Khon Kaen province in northeastern Thailand, took the 3,359 passing eggs, and scanned them all. Advanced periductal fibrosis — dense scarring around the bile ducts — showed on 23.6 percent of scans, more often in men. Eleven scans looked like cancer.

The risk factors for advanced fibrosis were infection intensity, as anyone would expect, and being younger than thirty, as nobody would (Mairiang 2012). It was cross-sectional — everybody measured once, nobody followed — so it cannot say who progressed, and selection into scanning can produce the same shape. But the comfortable sentence, that the scarring accrues gently with age, is not what the scans said.

Now the other half of the meal. Mitacek and colleagues (1999) ran more than 1,800 samples of fresh and preserved food from the four major regions of Thailand, collected 1988 to 1996, by gas chromatography. Nitrosamines — chemical carcinogens that form in some preserved, fermented and smoked foods — came up relatively high in fermented fish for three compounds at once: N-nitrosodimethylamine, N-nitrosopiperidine and N-nitrosopyrrolidine. In salted and dried fish, N-nitrosodimethylamine ran up to 66.5 micrograms per kilogram. The dish that carries the larva also carries, measurably, a chemical carcinogen.

It is food chemistry: it measures the dish, not any particular person. Case-control studies went a step further and asked people what they ate — pooled across eighteen lower-Mekong studies, eating high-nitrate foods was significantly associated with this cancer (Steele 2018). What nobody has done is measure one person's nitrosamine intake and follow that person to a cancer, which is why the dose is an inference.

Then the clock. In the population-based Khon Kaen Cancer Registry for 1985 to 2009, three-quarters of the bile duct cancers fell between fifty-five and sixty-nine years of age, men affected twice as often as women (Kamsa-ard 2011). The screening program for this disease opens at forty: Khuntikeo and colleagues (2015) report a cohort of mean age 53.8 years in which 89.1 percent had eaten uncooked fish and 42.2 percent of those tested for the fluke were infected.

Set the raw fish of a childhood beside a cancer clustering in the late fifties and sixties and you get four decades. Latency means the time between exposure and cancer, and this figure is not one: it is an inference from two cross-sectional pictures laid side by side. No cohort has followed infected people from a first infection to a cancer and reported the time between. The error has a name — reading an age distribution as a latency — and four decades is the right order of magnitude, not a measurement.

So what is it. Opisthorchis viverrini is a trematode — a fluke, one of the parasitic flatworms — in the family Opisthorchiidae, with two close relatives that belong here. Qian and colleagues (2024), in the current reference review, place Clonorchis sinensis in eastern Asia, O. viverrini in southeastern Asia and Opisthorchis felineus in eastern Europe, each cycling through people, animal reservoirs and two intermediate hosts, and each, in that review's careful wording, associated with liver and biliary complications, most importantly bile duct cancer. Associated. More than twelve million people in the Greater Mekong Subregion carry O. viverrini (Sripa 2021).

The cycle, in order. The adult lives in the bile ducts of a definitive host — the host in which it reaches sexual maturity, here a person, a cat or a dog — and its eggs leave in the feces (Buisson 2017). In fresh water the parasite passes through its first intermediate host, a small snail of the genus Bithynia, and emerges as a cercaria, the free-swimming larva, which burrows into a freshwater fish and settles in the flesh as a metacercaria, the dormant infective stage. Somebody eats that fish raw or lightly fermented and the cycle shuts in the bile duct.

Dao and colleagues (2017) counted both links at one endemic site in Binh Dinh Province, central Vietnam: 12,000 freshwater snails from six families and 754 fish of twelve species. Of the snails, 1,616 were Bithynia, of two kinds, and shedding of O. viverrini cercariae was found in those two and no other snail at all: 0.86 percent of the Bithynia siamensis goniomphalos and 0.14 percent of the Bithynia funiculata — rates within each species, not shares of the 12,000. Metacercariae turned up in ten of the twelve fish species: Carassius auratus, commonly eaten raw, at 74.0 percent, Rasbora aurotaenia at 55.8, Puntius brevis at 31.6. Under one percent in the snail, seventy-four percent in the fish people eat raw: the constriction is in the snail and the amplification is in the fish.

Two limits and one correction. The limits: one field survey, one district, one stretch of time, and prevalence rather than intensity — it cannot say how many larvae one infected fish carries, so it cannot say what dose a meal delivers. The correction: the standard line is that the carrier is a cyprinid, of the carp and minnow family, and in central Vietnam the metacercariae spanned both Cyprinidae and non-Cyprinidae. Naming the carp family alone takes the textbook's example for its boundary.

The Atlas page for this fluke holds the whole organism — genome, cryptic species, ultrasound findings, control programs — and this part will not repeat it. Two questions are left: how strong the case is that this animal causes human bile duct cancer, and what the gene-editing experiment showed. Neither can be weighed without first knowing how the worm arrives, how often, and what the cancer at the end of it has been counted at — so the dish and the registry come first.

The dish, the registry, and the curve that turned

Why thirty years of telling people not to eat raw fish moved so slowly has been measured. Wang and colleagues (2023) worked in two northeastern Thai villages, one riverside and one inland, asking about the two dishes the authors call the higher-risk ones: koi pla, a raw fish salad, and pla som, a fermented fish preparation. Buisson's review names such dishes as the vehicle for both the worm and the nitrosamines in the same mouthful (Buisson 2017). Men ate them on 64.5 days a year in one village and 40.4 in the other; women in the same villages, 4.1 and 4.3 days (Wang 2023).

The authors asked why, and the answer was not hunger. Consumption was driven mostly by what the paper calls cultural ecosystem services — the non-material things people get from the water and fields around them — the eating together, not the calories. Raw fish as a main source of protein, a provisioning service in the same vocabulary, accounted for less: 29.7 percent by the river, 16.1 percent inland.

Then the number that explains the campaigns. Taking part in sharing the dish significantly reduced the odds that a person would be willing to give it up, odds ratio 0.19 — an odds ratio being the odds of an outcome in one group divided by the odds in another, so willingness at about a fifth of the odds among non-sharers. A companion study in four villages in Kalasin Province adds the misinformation: embedded cultural practice was the reason 76.9 percent gave for eating raw fish, and only 15.3 percent named raw fish as the food that transmits the fluke (Wang 2021). The limits: small questionnaire samples, single rounds, and willingness to stop is a stated intention, not a behavior anyone watched.

There is one randomized test of the education itself. O'Connor and colleagues (2026) ran a cluster-randomized trial — schools randomized rather than children — of a cartoon package called The Magic Glasses Opisthorchiasis in Cambodia, Laos and Thailand. Against schools given the usual health education alone, knowledge and attitude scores rose by 19.2 and 25.3 percentage points in Cambodia and by 19.0 and 14.2 in Laos. In Thailand they fell, by 23.3 and 15.8. Behavior scores improved in no country, though parents in Cambodia and Laos reported better fish preparation; the endpoint was a questionnaire score, not an infection. A 2026 survey of 1,167 residents of Xinhui, in China, suggests why knowledge may be the wrong lever: what tracked with eating raw fish was refusal self-efficacy — whether a person believes they can say no when the dish comes round — rather than what they knew (Huang 2026).

So "the campaigns failed" is too blunt. Sripa and colleagues (2016) report infection approaching 85 percent prevalence in some endemic areas after more than thirty years of Thai control programs. Against that, nationally: measured Thai prevalence fell from over 15 percent in 1996 to 2.2 percent in 2019 (Sripa 2021). And in the same 2016 paper, an integrated program around Lawa Lake took the twelve villages there from about 60 percent infected to under 10, with infection in the cyprinid fish falling under 1 percent against a baseline maximum of 70, and no infected Bithynia snails found in recent field studies — an uncontrolled before-and-after by the program's own architects, no independent evaluation, no intervals. Attacking the whole life cycle in one lake basin did what leaflets and deworming days had not.

Now the disease. Cholangiocarcinoma is cancer of the cholangiocytes, the cells lining the bile ducts — a highly lethal adenocarcinoma, a cancer of gland-forming cells, classified by where it starts (Brindley 2021). Intrahepatic means inside the liver, between the smallest ductules and the second-order ducts, the definition in the joint 2023 guidelines of the European Association for the Study of the Liver and the International Liver Cancer Association. Perihilar means at the hilum, where the main ducts leave the liver. Distal means lower, in the common bile duct; perihilar and distal sit outside the liver, which is why they are grouped as extrahepatic, and Valle and colleagues (2021) take those three sites with gallbladder cancer as the biliary tract cancers. Most patients present with locally advanced or metastatic disease (Valle 2021).

It declares itself late, and the reason is structural: those 2023 European guidelines blame the silent presentation, the aggressiveness and the resistance to treatment. Banales and colleagues (2020) give the scale — about 15 percent of primary liver cancers, about 2 percent of cancer deaths worldwide each year — and state the diagnostic problem: noninvasive approaches are not accurate enough, so histological confirmation is necessary.

Khon Kaen's cancer registry was established in 1984 at Khon Kaen University, and Kamsa-ard and colleagues (2011) analyzed its first twenty-five years. Of 18,589 liver cancers from 1985 to 2009, 42 percent, 7,859, were hepatocellular carcinoma, the usual primary liver cancer, and 58 percent, 10,731, cholangiocarcinoma — the inverse of the pattern nearly everywhere else. The age-standardized rate, adjusted so populations of different ages compare fairly, was 44.3 per 100,000 men, 95 percent confidence interval — the range in which the true value probably lies — 38.9 to 49.7, and 17.6 per 100,000 women, 14.5 to 20.7. Single years ran to 62.0 in men, where the headline comes from.

Morphological verification — somebody looking down a microscope at cells or tissue from the tumor itself — was 10.8 percent: 1,141 cases out of 10,731. The rate quoted as the world's highest for this cancer rests on a registry in which nearly nine of every ten diagnoses were never confirmed by looking at the cancer. The same abstract adds a second caution: the rise across the first five or six years may be the registry becoming more complete rather than the disease more common. What verification is sits on the Atlas page on cancer pathology.

That is not simply carelessness. Talpur and colleagues (2026) pooled 38 studies and 4,912 patients with bile duct narrowings of uncertain cause, ranking five ways of taking tissue. Brushing cells from the narrowing came last on sensitivity, 45 percent — fewer than half the cancers — interval 38 to 52; a camera inside the duct, single-operator cholangioscopy, came first at 82 percent, 76 to 88. Specificity ran the other way: 97 percent for the brush, 89 for the camera. Those are credible intervals from a Bayesian model, not confidence intervals.

One more thing about "the highest in the world." As usually quoted it is an assertion in a review's preamble: Buisson (2017) writes that Khon Kaen has the highest incidence rate in the world, and offers no comparison. A second figure circulates for the same province, three times larger — 135.4 per 100,000 among males and 43.0 among females, in the screening program's own cohort profile (Khuntikeo 2015). It is not the registry's age-standardized rate for 1985 to 2009, the two cannot be reconciled from the abstracts, and this part uses the registry's own number while naming the bigger one out loud.

The best comparison that exists is Florio and colleagues (2020), from registries recording site and tissue type in 38 countries for the intrahepatic type and 33 for the extrahepatic, 1993 to 2012. The highest national rates were in Asia: South Korea 2.80 per 100,000 intrahepatic and 2.24 extrahepatic, Thailand 2.19 and 0.71. Do not divide 44.3 by any of those: a province is not a country, and an all-sites rate is not a subtype-specific one. No verified Western rate was found to set beside 44.3, so the familiar one or two per 100,000 is not said here.

Kamsa-Ard and colleagues (2021) took thirty years of the same registry, 1989 to 2018, and analyzed it three independent ways. The annual percent change — the average year-on-year movement in a rate — fell in both sexes: minus 3.1 percent a year in men, interval minus 4.0 to minus 2.1, and minus 2.4 in women, minus 3.6 to minus 1.2. Four other Thai registries agree for 2012 to 2021: average annual percent change minus 7.20 in men, minus 5.81 in women, across 6,379 cases (Sahat 2025). It is coming down.

It is coming down by birth cohort, which is what makes it credible. Men born in 1998 had an incidence rate ratio of 0.09 against men born in 1966, interval 0.07 to 0.12; for women 0.11, 0.07 to 0.17. An incidence rate ratio divides one group's rate by another's, so that is about an eleventh of the risk. The projection for 2028 is 7.6 per 100,000 in men and 3.6 in women. One caution: that abstract pairs the two rates with case counts that look transposed, so quote the rates, not the counts.

One error to name while we are here. Kamsa-ard and colleagues (2011) called the 1990 to 2009 decline of 0.7 percent a year in men significant, and printed an interval of minus 2.1 to plus 0.8, which includes no change at all. When a paper's own interval crosses zero, the interval wins.

Five-year relative survival in Khon Kaen was 10.9 percent, 10.3 to 11.6 (Kamsa-Ard 2021) — relative survival being the share still alive against people of the same age and sex in the general population. The figure usually quoted on air is under 5 percent — a review's number (Buisson 2017); the measured one is twice as high and still means nine in ten dead within five years. In the minority whose tumor can be cut out it is better: 63 resected intrahepatic cases in an endemic province gave median survival of 28.5 months and five-year survival of 30.1 percent (Tongsiri and Ingsirorat 2026). That gap, 30 percent against 11, is the argument for screening. The count behind the rates is about 20,000 deaths a year in Thailand alone (Taweesin 2025), and outside Thailand it is no better: 40,030 United States cases for 2001 to 2017, incidence rising 43.8 percent, median survival 8 months — a paper with three authors affiliated to a pharmaceutical company (Javle 2022).

The curve turned nonetheless: a generation born after 1990 is counted at a fraction of its grandfathers' rate. The epidemiology turned and the oncology did not.

How strong the link is, and where the damage is measured

So how strong is the link. Huang and colleagues (2024) screened 6,488 articles and pooled the twenty-two that qualified — cohort, case-control and cross-sectional together, 34,367 participants. Infection with Clonorchis sinensis, Opisthorchis viverrini or Opisthorchis felineus carried a cholangiocarcinoma odds ratio of 4.24, 95 percent confidence interval 3.33 to 5.39. By species, each with its own interval: 4.49 for C. sinensis, 3.43 to 5.87, and 3.69 for O. viverrini, 2.07 to 6.55. The estimate for this part's own fluke is the less precise of the two, and its lower bound is close to twofold.

An odds ratio measures association — how much more often an exposure turns up among people who have the disease than among people who do not — not how much more likely you are to get the disease; and an interval crossing 1 is not statistically significant.

Fourfold is serious, and it does not say that carrying the worm gives you cancer; most people who carry these flukes will not get bile duct cancer. The same analysis pooled risk ratios too — the measure that does compare rates of new disease between exposed and unexposed — and got 10.43, interval 2.90 to 37.47. A spread that wide tells you its underlying studies disagree.

Homsana and colleagues (2024) worked through 3,400 people in high-endemic rural communities in southern Laos: abdominal ultrasound on all of them, blood tests for hepatitis B and diabetes, and stool examination by the Kato-Katz technique — a thick smear of stool read under a microscope, named for the two people who devised it, the standard field count for these infections, hunting for an egg a fortieth of a millimeter long, lidded, its shell patterned like melon rind (Sukontason 1999); its limits become the point later. Of the 3,400, 88.3 percent were infected with O. viverrini, and prevalence of suspected cholangiocarcinoma was 7.2 percent, 95 percent confidence interval 5.4 to 9.6, rising with age and higher in men at every age. In the multivariable analysis, suspected cancer was associated with fluke infection at an adjusted odds ratio of 3.4 (1.7 to 6.5) — adjusted meaning other measured factors were accounted for — and with gallbladder removal at 2.7 (1.5 to 4.9).

The endpoint is suspected cholangiocarcinoma: a lesion that looked like cancer on an ultrasound screen. That is a surrogate endpoint, and surrogates can move without the real thing following — no technicality where dense scarring and early tumor look alike. And at 88.3 percent infected, the uninfected comparison group was fewer than one person in eight.

The scale underneath all of it: the three flukes together infect approximately 25 million people worldwide, a figure both the pooled analysis and the drug-efficacy meta-analysis of Qian and colleagues (2022) open on. Two of the three — O. viverrini and C. sinensis — are in the International Agency for Research on Cancer's Group 1, the category for agents whose evidence of causing cancer in people is held sufficient rather than suspected; only three parasites are in it at all (Kim 2016). How that agency reaches a category is on the Atlas page about it.

The current consensus statement is Smout and colleagues (2024), which sorts the mechanism into three strands. The flukes secrete growth factors, digestive enzymes and extracellular vesicles — small membrane-wrapped packages cells release and other cells take up — and those secretions drive abnormal cell development in the ducts where the worms live; they combine with aberrant inflammation; and both combine with repeated cycles of chronic wounding where the parasite attaches and grazes on the epithelium. Two notes before leaning on that review: it is a characterization, not a measurement, and its authors are the laboratory group whose own candidate molecule the next section is about.

Two things about the mechanical strand get overstated, and it is the weakest of the three. The first is anatomical precision: the reviews say only what was just quoted, attachment and grazing, and no source I opened pins the injury on the oral sucker. The second is quantity. Sithithaworn and colleagues (1991) recovered worms at autopsy from 181 cases in northeastern Thailand, at burdens low enough that stool examination missed twenty cases carrying fewer than twenty worms each. Thousands of flukes in a duct is unsourced.

What does exist is a bound. Thamavit and colleagues (1993) gave Syrian hamsters a single intraperitoneal injection of dimethylnitrosamine at 20 milligrams per kilogram, then two weeks later either a sham operation or surgical ligation of the bile duct draining one liver lobe — a tie around the duct, no parasite anywhere in the experiment. At week 40 the initiated-and-ligated animals had cholangiocarcinoma in 39.1 percent, cholangiofibrosis — duct proliferation knotted with scarring — in 60.9 percent, and mucous cystadenomas in 21.7 percent; the nitrosamine alone gave cholangiofibrosis in 5 percent and no tumors, ligation alone nothing. So physical injury, in an animal already primed with a chemical carcinogen, drives this tumor in four animals in ten: the worm is not required for the mechanism to work. But a surgical tie is complete, permanent obstruction, a far cruder insult than a flatworm browsing a surface. It brackets the mechanical strand without measuring the parasite's share.

Pinlaor and colleagues (2004) infected hamsters once, twice or three times with O. viverrini and looked for two chemical injuries to DNA: 8-oxo-deoxyguanosine, shortened to 8-oxodG, a guanine base oxygen damage has altered, and 8-nitroguanine, a guanine base nitrogen-based reactive molecules have altered. In the liver, 8-oxodG peaked on day 3 in the thrice-infected animals, day 7 in the twice-infected, day 21 in the once-infected; in the duct lining both rose in that order, and after the inflammatory cells had receded. The authors' explanation is inducible nitric oxide synthase, an enzyme inflamed tissue switches on to make nitric oxide: by day 90 the lining was expressing it itself, in the same dose order. Two years later the same group gave infected hamsters a week of praziquantel and found it almost completely abolished both lesions in the lining cells and reduced inducible nitric oxide synthase and NF-kappa B, a master switch for inflammatory genes, even though worms were still present (Pinlaor 2006). DNA lesions reduced in hamsters is not cancers prevented in people.

Kawanishi and colleagues (2017) report 8-nitroguanine strongly formed, by way of inducible nitric oxide synthase, at the cancer sites of patients across the whole set of infectious agents rated proven human carcinogens. In fluke-associated cholangiocarcinoma it was concentrated in cells positive for Oct3/4 and CD133, markers for stem cells — the self-renewing cells a tissue regrows from, so damage a whole lineage can inherit. Two limits: descriptive staining, no counts, no effect sizes, in a review the authors wrote about their own work; and the tissue is established cancer, which cannot say whether the lesion arrived before the tumor or after.

Interleukin-6 is a signaling protein cells release to coordinate inflammation. Sripa and colleagues (2012) found it 58 times higher in plasma from people with advanced periductal fibrosis and 221 times higher in people with fluke-associated bile duct cancer than in matched controls; adjusted for age and sex, the highest quartile carried 19 times the risk of advanced fibrosis and 150 times the risk of cancer against those with none detectable. But the authors state, in the same abstract, that high circulating interleukin-6 is not related to infection with O. viverrini but to the development of the advanced and often lethal pathology chronic infection produces. So 221-fold is not a measure of what the worm does; it marks whose tissue has already gone wrong. What is wrong, then, is reading that plasma figure as the parasite commandeering host interleukin-6 — the hijack error. What is not wrong is the idea that the worm can raise the cytokine at all: in culture it does, in both species anybody has examined (He 2023; Hembasat 2023). The second error is calling this a cytokine storm, which names an acute, systemic, often fatal surge, not four decades of low-grade inflammation in a duct. The companion study makes the point with a denominator: among 328 infected people, interleukin-6 responses to fluke secretions were 8 times higher in the 200 with advanced fibrosis than in the 128 without (Sripa 2009). Both groups had the worm.

Heneberg (2026) notes that helminth infection is among the commonest causes of chronic inflammation in human beings and that only a small subset is linked to cancer at all, which undercuts the assumption that chronic inflammation is intrinsically carcinogenic. His threshold proposal — tumors emerge only where sustained inflammation, long exposure, genotoxic stress, a vulnerable tissue and permissive cofactors converge — is a hypothesis in a single-author review. One thing it names has a measured instance: hamsters carrying O. viverrini with a cagA-positive Helicobacter pylori, a bacterium the fluke harbors in its gut, developed worse inflammation and earlier precancerous change than animals with the fluke alone (Suyapoh 2020).

One gene, two experiments, and the limits of both

The third strand is a molecule the worm makes on purpose, and it starts with a word this show has used loosely. Granulin is not one protein but a module: a short stretch of protein chain, rich in cysteine, held in shape by disulfide crosslinks. The proteins carrying it are chains of those modules; the full chain is progranulin, and enzymes cut it into separate granulin peptides of about six kilodaltons.

In 2009 Smout and colleagues ran a proteomic screen on everything adult Opisthorchis viverrini releases into its surroundings and found a homolog of human granulin — a protein descended from the same ancestral gene in another species. They named it Ov-GRN-1: a single granulin domain of about nine kilodaltons, made most strongly in the worm's gut and tegument, the living skin that covers a fluke, and found sitting in place on the bile duct lining cells of infected hamsters (Smout 2009; Arunsan 2019).

Ov-GRN-1 grown in Escherichia coli and refolded made mouse fibroblasts divide at nanomolar concentrations — a very low dose for a protein secreted into a bile duct. The proliferation was blocked by U0126, which inhibits the enzyme switching on the mitogen-activated protein kinase pathway, one of the signal chains a cell uses to decide whether to divide. Antibodies against the protein blocked the whole secreted mixture's proliferative power, which makes Ov-GRN-1 the major growth factor in that mixture rather than one of many (Smout 2009). Two years later the same group made it as a clean single molecule, and it still drove proliferation at nanomolar doses in human bile duct cell lines (Smout 2011).

It also closes wounds and builds blood vessels. Secretions from gene-silenced flukes were markedly worse at closing a scratch wound in human bile duct cells, and the protein alone sped skin-wound healing in mice and grew new vessels in a quail egg membrane (Smout 2015); cells from the lining of human umbilical vein were significantly stimulated at ten and forty nanomolar (Haugen 2018). Cells in dishes, mouse skin, a bird embryo: nobody has measured this protein in a person.

And here is what makes granulin interesting rather than ordinary: no study in this literature reports it damaging DNA, and none appears to have tested it for that. The fair reading is not that it has been checked and cleared as a mutagen but that the genotoxicity experiment has not been done, and that what it contributes runs through proliferation rather than mutation. It keeps telling cells to divide, for decades, in tissue already being injured by other means. Two claims in circulation have to go. One is that the 2009 study used RNA interference to silence the gene, and that silenced worms could no longer make human cells grow. It did not; 2009 was proteomics plus antibody neutralization. RNA interference arrived five years later: Papatpremsiri and colleagues (2014) cut the message by 95 percent at day three, and co-culture with silenced flukes cut overgrowth 25 percent in a normal cholangiocyte line and 92 percent in a cancer line.

The second claim is that Ov-GRN-1 binds the epidermal growth factor receptor. No paper opened for this episode reports that, and no receptor for the parasite protein has been identified at all. The likely origin is the U0126 result, because the enzyme that drug blocks sits downstream of many receptors, and a downstream inhibitor is not a binding partner. The discovery paper is also commonly miscited: wrong first author, wrong year, and an identifier that resolves to a paper about crown and root rot in wheat. One error of the same family: the fluke's genome was published in 2014 as a 634.5-megabase draft (Young 2014), no chromosome address for Ov-grn-1 has ever been published, and any citation supplying one is a fabricated genomic address — a coordinate invented for a real gene.

Now the experiment. In 2019 Arunsan and colleagues did something not done before in any flatworm: programmed gene editing. The tool was CRISPR/Cas9 — a guide of RNA about twenty letters long that matches a chosen stretch of DNA, and an enzyme, Cas9, that cuts where the guide lands, after which the cell's error-prone repair usually leaves small insertions or deletions that wreck the gene. A gene wrecked that way is called a knockout, and so, loosely, are the worms and animals carrying it. The target was the first coding stretch of Ov-grn-1, the fluke's own granulin gene, inside living flukes: nothing in the hamster's genome was touched and no human gene was involved. A plasmid — a small ring of DNA — carrying Cas9 and the guide was pushed in by electroporation, one square-wave pulse of 125 volts for 20 milliseconds. The metacercariae could not be edited, because the cyst wall blocks it; the juveniles newly excysted from them could.

Thirty male Syrian golden hamsters — the standard animal, because it develops bile duct cancer the way infected people do — were split into three groups of ten: uninfected, wild-type flukes, edited flukes, a hundred juveniles each, five per group examined at days 14 and 60. The edited flukes colonized the bile ducts and matured into adults, so the result is not simply sick worms causing less disease — though the day-60 counts come from two livers per group. At fourteen days the duct lining in wild-type infection was thickened to roughly 500 percent of uninfected, and in the knockout infection to 145 percent; by sixty days, 216 against 162. Collagen was roughly a quarter lower in the knockout group — the text says 28 percent, the figure legend 23 — and ACTA2, the gene for the actin scar-forming cells make, was 94 percent lower by median.

What that 2019 experiment did not do matters as much. There was no chemical carcinogen in it and there were no tumors. Thickening and scarring are surrogate endpoints: hyperplasia is tissue thickened by cells multiplying while the cells still look normal, and both it and fibrosis are risk states, not cancer. The design was thinner than usually described, too — no arm carrying a guide aimed at nothing, a control the authors themselves recommend for future work.

Chaiyadet and colleagues (2022) ran the same parasite-gene knockout and added dimethylnitrosamine to the drinking water at 12.5 parts per million from week two to week ten, then examined the animals after death at week 24. Cholangiocarcinoma appeared in 10 of 12 control hamsters, 8 high grade — three or more tumor clusters per microscope field — against 7 of 13 knockout hamsters, 6 high grade. Duct cells staining for mutant p53, the damaged form of the cell's main brake on uncontrolled division, fell from 61.1 percent to 7.5 percent, 95 percent confidence intervals 43.6 to 68.3 and 2.3 to 27.5. Cells caught dividing beside a fluke fell 4.8-fold, from 15.0 to 3.1 percent. Scarring on the Ishak scale, zero to six, had a median of 4 in controls and 2 in the knockouts.

One correction belongs beside that stain, because this show attached its human counterpart to the wrong paper. The percentages usually quoted — TP53 damaged in 44.4 percent of cases, KRAS and SMAD4 in 16.7 each, GNAS in 9.3 — come from an exome study of eight fluke-associated tumors with matched normal tissue, published in Nature Genetics in 2012 and then screened for prevalence across 46 further cases (Ong 2012). They are not from the 2017 whole-genome paper, which is in Cancer Discovery, took 489 cholangiocarcinomas from ten countries, and put the fluke-positive ones in two of its four clusters (Jusakul 2017). Eight tumors is a thin base, but it gives the hamster number its human echo.

The editing was mosaic: only some cells in a many-celled worm take up the plasmid, so every animal is a patchwork. In the adult flukes sequenced from the cancer experiment the median worm carried insertions or deletions at the target site in 3.1 percent of gene copies read, ranging from none at all to 91 percent. What collapsed reliably was not the gene but the message: about 11 percent of wild type in adult knockout worms (Chaiyadet 2022). The explanation came from the same laboratory four years later — the plasmid reaches only the nuclei of the tegumental syncytium, the worm's surface layer, a single sheet of cytoplasm with many nuclei and no dividing walls, where these genes are switched on (Chaiyadet 2026). No heritable knockout line exists in any of these flukes, which the laboratory that built the method says in print (Ittiprasert and Brindley 2024).

The second limit is worm burden. In the cancer study's separate fourteen-week arm, 38.9 percent fewer knockout worms were recovered, 36.7 against 60.0 per hamster, which the authors call a fitness deficit, qualifying the 2019 claim that edited flukes develop normally. In the cancer arm they were never counted; the livers had to be preserved whole for sectioning. Two things narrow that. The authors measured scarring and cell division immediately around individual flukes, worm for worm. And the study's other knockout group lost even more worms, 64.5 percent fewer, yet showed no significant softening of fibrosis or mutant p53: cancer in 9 of 13 hamsters, mutant p53 in 67.7 percent of duct cells, slightly above the controls, Ishak median 3 against the controls' 4 and not significantly different. Say the half that cuts the other way too: only four of those nine cancers were high grade, against eight of twelve in controls.

That other group is the third limit. The intended comparator was a second gene the worm uses to do harm, the surface tetraspanin Ov-tsp-2, molecular glue on the worm's skin. Knocking it out proved lethal: those worms survived in far smaller numbers, and the survivors carried near-normal message levels, implying only the unedited ones made it. The authors write in print that the choice was, in hindsight, a sub-optimal control (Chaiyadet 2022). The reason surfaced in 2026: editing that gene leaves the tegument eroded and pitted, its area down 40 to 47 percent, so the worm's protective surface fails to form — and the same paper states flatly that knocking out granulin is not lethal in a living host (Chaiyadet 2026). That retires an older belief: the 2014 silencing work had suggested granulin might be essential for adult survival in culture.

The fourth limit: no outside laboratory has replicated the phenotype, and nearly every primary paper in this line shares a core author set. The only genuinely independent voice on the editing work is a commentary in the same journal by two parasitologists at Queen's University Belfast, and it adds no data (McVeigh and Maule 2019). The nearest corroboration is the study quoted earlier on interleukin-6, from a separate group working on the sister fluke (He 2023): a different species, cells and mice, no cancer endpoint.

And the worm has more than one granulin gene. A second single-domain granulin sits in its genome (Smout 2015), and so does a full fluke progranulin, Ov-PGRN: 846 amino acids, a predicted 91.61 kilodaltons, one half plus seven complete granulin domains, detected by western blot in both the worm's body and its secretions and localized to tegument and parenchyma, with a peptide from it driving cholangiocyte proliferation and raising interleukin-6 and interleukin-8 (Hembasat 2023, a preprint — posted before peer review). Deleting Ov-grn-1 deletes neither. The non-detection worth stating is a different one: Ov-GRN-1 itself is not detected in the worm's extracellular vesicles (Chaiyadet 2026).

Removing one secreted parasite gene reduced the fibrosis, the proliferation, the mutant p53 and the number of high-grade tumors. Seven of thirteen knockout hamsters still developed cholangiocarcinoma, the same kind in the same places. Granulin is contributory, not sufficient — a parasite's gene with a human cancer pathway downstream of it, which is a large claim without being a total one. The honest headline is reduced malignant transformation, not prevention.

One more experiment has to be separated out loud. In 2020 the same group did knock out a human gene, progranulin, the host's own, with CRISPR/Cas9 — in a cultured human bile duct cell line, more than 80 percent down in message and protein, to clear the host's progranulin out of the way so the parasite protein could be read cleanly (Arunsan 2020). No animal was edited, no parasite was edited, nothing about disease was measured. Parasite gene edited in a living worm: 2019 and 2022. Host gene edited in a dish: 2020. Any sentence that merges them is wrong.

The reason that control was needed is biological: the host's own granulin rises during this carcinogenesis too. Upontain and colleagues (2018) ran 120 Syrian golden hamsters in four groups of thirty and found mammalian granulin raised in the duct-lining cells of cancer lesions in the animals given the nitrosamine, with and without the fluke — while the proliferative lesions themselves were worst in the animals given both. So the chemical raises host granulin on its own. But two granulins are present in one bile duct and both go up, which is why the parasite's could not be read in a cell line until the host's was deleted.

No PubMed record for these papers carries a conflict-of-interest declaration, so nothing is verified either way; every affiliation is a university, public institute or government agency. What the authors state is a translational interest in the same molecule: the James Cook University group is developing peptides built on the shape of Ov-GRN-1 as a wound-healing drug, potent in mouse skin wounds (Bansal 2017), and was still producing them in 2025 (Dastpeyman 2025). The protein argued to help cause a bile duct cancer is also a drug lead for chronic ulcers, which sharpens the interest rather than softening it. The Atlas page on granulin carries the module's evolution, the human diseases of losing it, and the fight over which receptor progranulin binds.

The other two flukes, the drug, and the clock

There are two more flukes. Clonorchis sinensis, the Chinese liver fluke, is endemic across Korea, China, Taiwan, Vietnam and far eastern Russia, and still actively transmitted (Kim 2016). In Korea's eighth National Survey of Intestinal Parasitic Infections, 2012, it was the most prevalent parasite of any species surveyed in the general population, at 1.86 percent — almost two percent of a wealthy, universally insured country carrying a Group 1 carcinogen in their bile ducts. It has not always been there: O. viverrini was put in Group 1 in 1994 and C. sinensis only in 2009, fifteen years later (Buisson 2017; Shin 2010). Which category it held in between is not established in anything read here, so the line about moving up from Group 2A is not said.

Shin and colleagues (2010), writing from that agency, put the Korean arithmetic on the record. Stool egg positivity in endemic areas was above 25 percent of adults tested in 2005 to 2008. Pooling the case-control literature then available gave a summary odds ratio of 4.7, interval 2.2 to 9.8. And approximately 10 percent of all cholangiocarcinomas in Korea, they estimated, were caused by chronic infection with this fluke: a population attributable fraction, meaning the share of cases that would not have happened without the exposure, calculated from a survey-based prevalence estimate rather than counted from patients. The 2024 pooled analysis gives 4.49. Two independent pools, fourteen years apart, nearly the same place.

That same paper points the opposite way from the Thai curve: across 1999 to 2005, cholangiocarcinoma incidence in Korea was rising — 8 percent a year in men, 11 percent in women — while primary liver cancer was falling. And transmission continues in five major Korean river basins, the fish still carrying cysts (Yoo 2022). The largest reservoir is not Korea: in China about 9.46 million were infected in 2020, nearly 100 million counted at risk (Fang 2026, a review in Chinese read here in English abstract only). The Atlas page on Clonorchis carries the survey history.

The third fluke is Opisthorchis felineus, the cat liver fluke of Russia and eastern Europe, the honest hole in this field. Four things are commonly said about it and all four need work. First, the prevalence. The circulating figure is 80 percent, and it is traceable: it sits in the opening lines of a 2015 paper by Maksimova and colleagues in the Russian journal Parazitologiia, as the flat assertion that infestation of the endemic-area population reaches 80 percent. Nothing is attached to it. The measurement is elsewhere: Fedorova and colleagues (2020) enrolled 600 household members aged seven and over in the rural Shegarskiy district of Tomsk Oblast and got 488 through two stool examinations. Prevalence: 60.2 percent. Not 80. The error has a name — an assertion in a review's preamble, repeated until it sounds like data.

Second, the hamster experiment. You will hear that fluke plus a nitrosamine produced bile duct cancer in every single animal by week eighteen. That is true, and it is the wrong species. The week-eighteen result is from that same Parazitologiia paper: golden hamsters in four groups, twenty animals in each treated group, dimethylnitrosamine at 12.5 parts per million, fifty O. felineus metacercariae, and cholangiocarcinoma at week eighteen in all the animals given both. Flukes alone produced hyperplasia and dysplasia in the duct lining, which the authors read as precancerous. So the number belongs to O. felineus, with its limits: Russian-language with an English abstract, no digital object identifier, twenty animals a group, not one confidence interval. The group's English companion scored the histology from ten to thirty weeks and gave no incidence percentage (Maksimova 2015).

Third, the human epidemiology that was not there. Fedorova and colleagues (2016) reviewed official surveillance returns across all 83 political and geographical units of the Russian Federation for 2011 to 2013. Reported infection averaged 24.7 cases per 100,000 a year, reaching 599.7 in the Khanty-Mansiysk district and zero in 27 of the 83 units; liver and bile duct cancer ran at 4.8 per 100,000. The two did not move together — correlation coefficient 0.20, p-value 0.07 — and the abstract says flatly that to date no association has been demonstrated. But this is an ecological study, territories compared against each other with nobody's infection matched to anybody's tumor, the weakest observational design in use.

Fourth, the reversal. Fedorova and colleagues (2023) ran a hospital-based case-control study in Tomsk Oblast and Khanty-Mansiysk, 2017 to 2020: 40 patients with histologically confirmed cholangiocarcinoma, each matched to four hospital controls on age, sex and residence. Exposure to O. felineus gave an odds ratio of 3.9, interval 1.4 to 10.8, p-value .008; eating river fish gave 5.5 and fishing 3.3. Forty cases is small, the interval wide, the controls hospital patients, part of the exposure history self-reported, and this is association. But the estimate sits in the same range as the pooled figures for the two flukes in Group 1. Anyone still saying this fluke has never been linked to cancer in people is out of date, and has been since 2023.

The classification question now has a dated answer. A review published in January 2026 states that this fluke currently sits in Group 3 — the category for agents the evidence does not let the agency classify either way — and argues for moving it up on the strength of that 2023 result (Jain 2026). The same review supplies the mechanism this section otherwise lacks: interference with DNA excision repair, reactive oxygen and nitrogen species, by-products of lipid peroxidation, oxysterol-like compounds. The authors of the 2015 hamster paper wrote that their own result might be used to define this fluke's group at all. No Monograph has been read at source here, and that caveat stands. And the fluke is now diagnosed in western Europe: a 2026 report from Bonn describes a case in Germany, where no certified serology exists, diagnosis rests on repeated stool microscopy, and autochthonous cases are now reported inside the European Union (Dirik 2026).

So treat it. Praziquantel kills these worms, and the honest version of how well comes from a network meta-analysis by Qian and colleagues (2022) — 26 trials from 25 studies, 3,340 participants, pooling trials that did not all test the same pairs. For O. viverrini, the highest predicted cure rate was 93.8 percent, interval 85.7 to 97.5, for 50 milligrams per kilogram followed by 25 more in a single day; a single 50-milligram dose gave 92.1 percent on a much looser 64.9 to 98.6. Of the 26 trials, eighteen were in C. sinensis, seven in O. viverrini and one in O. felineus, so that 93.8 percent rests on seven trials, graded low quality along with most of the others, especially those before 2000. Cure rate means eggs gone from the stool.

Killing the worm is not the same as undoing what it did. Wangboon and colleagues (2018) followed Khon Kaen participants before praziquantel and one year after, with ultrasound and a urinary marker of oxidative DNA damage. In the 14 infected people with no periductal fibrosis to begin with, the marker fell significantly; in the 52 who had fibrosis it did not move at all, and 34.6 percent of those 52 still had the fibrosis at both scans. About a third of treated people with established scarring keep it, and once the scarring exists the DNA-damage signature stops answering the drug. Fourteen people and fifty-two, one year, a scan lesion for a cancer.

Then the conflict, which this part will stage rather than settle. Steele and colleagues (2018) meta-analyzed 17 risk factors across 18 lower-Mekong studies and found praziquantel treatment among those significantly associated with higher cholangiocarcinoma risk. That signal is almost certainly confounding: the most-treated are the most-infected and the most-exposed, and an observational design cannot pull them apart. The same analysis holds the strongest single sentence against overstating the parasite's share. Its largest effect was not the fluke: alcohol combined with smoking gave an odds ratio of 11.1, interval 5.63 to 21.92, and the authors state in print that the combination is a greater risk factor for this cancer than O. viverrini exposure itself.

Pointing the other way on the drug, Intajarurnsan and colleagues (2016) ultrasound-screened 55,246 people in the Cholangiocarcinoma Screening and Care Program, found periductal fibrosis in 33.0 percent, and found more than two praziquantel treatments carrying an adjusted odds ratio of 0.54 for that fibrosis, interval 0.47 to 0.63. But diabetes came out just as protective in the same model, at 0.57 — and in the hamster diabetes does the opposite: worm establishment rates track fasting plasma glucose, and worms from diabetic hosts turn up their virulence genes, granulin and the tetraspanins among them (Chaidee 2024). That 0.57 points at the model, not at biology, and the authors flag their own inverse associations as needing confirmation.

Meanwhile the Khon Kaen parasitology group ties the fall in incidence to decades of mass drug administration — treating whole communities on schedule rather than only those who test positive — and health education. That is a review's discussion, with no numbers in the abstract to check (Sithithaworn 2014). And Buisson (2017) keeps the uncomfortable alternative alive as a hypothesis and nothing more: that alternating cure and reinfection may itself promote carcinogenesis. The hypothesis now has a measured reinfection rate attached to it, if not an endpoint: 63.7 per 100 person-years after treatment, in the cohort coming next. Both positions cannot be straightforwardly true, and the trial that would decide is not going to be run. Crellen and colleagues (2025) say why in print: a randomized trial of deworming with a biliary-cancer endpoint could be considered unethical, because the control arm would be denied a cheap, licensed, effective drug for decades.

One thing goes in before the clock, because it measures the instrument everything after this rests on. Kopolrat and colleagues (2026) followed 612 people in northeast Thailand for 24 weeks, testing two ways at once. Baseline prevalence was 8.1 percent by stool concentration and 41.0 percent by a urine antigen assay — five times higher. Incidence was 7.5 per 100 person-years by stool against 64.6 by urine, and reinfection after praziquantel 5.9 against 63.7: a tenfold gap. Raw fish in the previous six months carried an adjusted risk ratio of 7.52, and previous praziquantel treatment was itself a risk factor at 1.49 — prospective this time, and still observational. Stool microscopy is the instrument behind every survey the falling curve is built from. This does not overturn the birth-cohort fall, too large to be an assay artifact. It does put an error bar on "zero worms by thirty."

Which brings the clock. Crellen and colleagues (2025) took 22 paired tumor and normal whole genomes from bile duct cancer patients in northeast Thailand, all born before 1980, and in the 17 with enough tumor content timed the chromosomal amplifications under their driver genes — a driver gene being one whose damage pushes a cell toward cancer, an amplification an extra copy of the stretch carrying it. The earliest such amplification in each tumor dated to a median age of 30, interquartile range 20 to 43: two to four decades before diagnosis. The dating assumes a mutation type that accumulates at a clock-like rate, and the authors test it — if the rate accelerates fivefold across a lifetime, the figure moves to 36 on average, interquartile range 28 to 46. They then checked the clock against 10,737 cholangiocarcinoma cases diagnosed in Khon Kaen between 1985 and 2009, six more than the registry's own published total for that window and the same series counted slightly differently, and got an induction period of 28 years and a latent period of 32.

Against that, the worm's own span: 13 years inside a human host, 90 percent credible interval 7 to 26 — a credible interval being the range a model of this kind holds the value most likely to occupy. That is the only measured estimate of the lifespan in this corpus, and it sits below the ten-to-thirty years the opening quoted from a review; the authors ask that the interval be used rather than the point estimate. For people born between 1960 and 1989 in endemic Thailand, first infection came at 2.2 years of age, 88 percent were exposed at some point, and the median person had cumulatively acquired 72 worms by 30. The heaviest-infected tenth: 691.

Then the control programs, and the number this part ends on. Refitting the model to the surveys taken afterward, the age-dependent transmission rate had fallen almost fortyfold. First infection moved from 2.2 years to 8.5, the share ever exposed from 88 percent to 32. And the median cumulative worm burden by age 30, for people born after 1990, is zero, the model's interval around it zero to zero; in the heaviest tenth it is ten worms, down from 691. These are model outputs fitted to 11,517 survey participants across 27 years, not worms counted in anybody. The same paper puts the lifetime probability of a cholangiocarcinoma diagnosis at 4.9 percent given exposure against 0.34 percent without it, fourteenfold, and attributes 92 percent of cases diagnosed there between 1997 and 2009 to the parasite.

Hold both halves. The attributable fraction is enormous, and the exposure driving it has largely gone. The lag is the cruel part: amplifications laid down at thirty, in people infected at two, will keep declaring themselves as cancer for decades after the last worm, which is why the curve falls slowly and screening still matters. One honest limit on the ending: the curve that turned is Thailand's. The other fluke in Group 1 sits in a country where, on the last published trend, this cancer was going the other way. The Atlas page on cholangiocarcinoma carries the registry series, the anatomical subtypes and the survival figures. An episode about parasites and cancer ends more credibly on a falling curve than on a warning, and this one gets to.

Sources

Every paper referenced on air, in the order it comes up. Links go to the publisher via DOI.