What Reaches the Glass

Part four follows a specimen from the body to the slide and counts what each step costs. It opens in Laos, on lung fluke eggs on six of 263 tuberculosis slides somebody had already read, four of their five patients never diagnosed. Then the chain, stage by stage: the minutes before fixative, the bench where one person decides which slices get looked at, the jar, the acid that bone goes into, the wax and the blade. Then the specimens that never become a block at all: needle aspirates, spun fluids, spinal fluid, marrow, a drop of blood, and the body after death. Ten percent formalin is not ten percent formaldehyde. Underfixation, not overfixation, was the problem when somebody tested it. And a worm in a wax block keeps its cuticle and its DNA, though fixation trims what a DNA test can find and long heating in staining costs eggs. What the chain does is choose: how many pieces, how much fluid, which hour of the day, what the reader was asked to find. Then the stain and the reader, the specimen-by-specimen head-to-heads, the misreadings that run the other way, and the worms that were found when they did reach the glass.

Show notes

Start in Laos, in 2009 and 2010. A team went back to tuberculosis slide archives in two provinces, Attapeu in the far south and Luang Namtha in the north, and pulled 263 sputum smears, sputum being what a person coughs up from the lungs, every one already stained and read. The stain was Ziehl-Neelsen, the carbol fuchsin method that turns the tuberculosis bacterium red against a blue counterstain; carbol is phenol, which is what that syllable means.

On six there were fluke eggs. Slesak and colleagues (2011) report Paragonimus eggs on 6 of the 263, from five patients: four in the south never diagnosed with it, one in the north who already had been. Paragonimus is the lung fluke, a flatworm caught from undercooked freshwater crab and crayfish, which settles in the lungs and lays eggs that are coughed up. It brings chronic cough, chest pain and blood in the sputum: a patient who looks like a tuberculosis patient. The eggs are brownish-reddish and enormous beside a bacterium. The Atlas page on the lung fluke carries its life cycle and the cases where it passed for cancer.

Hold what that is and is not. For fifty years, the authors write, this stain had been believed to destroy those eggs. It had not: across the study fewer than 19 percent were even fragmented. They sat in a trained technologist's field of view, and for four of those five patients the report answered only the question it had been asked. The whole failure is that the question was tuberculosis.

Two more details from the paper are the method of this entire part. Slides kept steaming five minutes during staining held 29 eggs per slide on average against 42 on briefly heated ones, p of 0.01: if heating made no difference, a gap this big would turn up about one time in a hundred. The authors read it as heat degenerating the egg wall: a staining step that costs about three eggs in ten, by my arithmetic, when it runs long, though current practice heats briefly. And smears from the bloodstained part of the sputum held 3.3 eggs per slide against 0.7 from the clear part, p of 0.016: which part a human chose to smear.

Its weight, honestly: an archival re-read by people who had decided to look, 263 slides, six positives. It shows that this happens, with nothing broken, not how often it happens elsewhere.

Here is the claim this hour tests, as it reached the show: “The pipeline destroys the evidence. Formalin kills and dissolves whatever is in the jar, so a worm inside a tumor is gone before anyone could look.” And the question under it: between the body and the slide, where can a parasite be lost? Destroyed by the chemistry, or simply not cut, not spun, not smeared, not recognized?

The thesis, up front. The route from a patient to a diagnosis is a chain of lossy steps, and almost every loss has been measured, usually by the field itself. Ranked as I read those measurements, and the ranking is mine since nothing puts them on one scale, the dominant loss is not chemistry. It is sampling and attention: what got cut, what got mounted, what the reader had been asked to find.

The chain has eleven stages.

One. Cold ischemic time, the minutes between tissue losing its blood supply and reaching fixative. Two. The bench, which the trade calls grossing. Three. Fixation. Four. Decalcification, only if there is bone. Five. Dehydration, clearing and paraffin infiltration.

Six. Embedding. Seven. Microtomy, the blade. Eight. Flotation and mounting. Nine. Staining. Ten. A human being looks at the slide. Eleven. The tests that exist only if somebody orders them: antibodies, gene panels, chromosomes. Part five takes that stage; this episode follows the specimen as far as the slide.

This episode reads aloud only the chemistry the claim depends on; the Atlas page on formalin fixation carries the rest, and the page on how cancer is tested, the specimen matrix. The page on iodine in the specimen takes apart a related claim, that iodine on the skin or on the tissue kills the evidence, which part five answers on air.

The frame, so nothing later sounds like a reveal: no fixative or processing step in those eleven stages has been shown to dissolve a worm, though long heating in staining costs eggs. The chain exists to stop tissue decaying and make it hard enough to slice, and a parasite in it is fixed and hardened with everything else. Nagayasu and colleagues (2026) published a case that shows it, a worm still readable in a fixed lung nodule, and it comes later.

One piece of honesty first. I found no study that split specimens of known worm content, processed half routinely and had blinded readers score recognition. So “nothing in the chain dissolves a worm” is an inference: from the chemistry, from cases like that one, from the chain's purpose being preservation. What the chain does, measurably, is decide what gets looked at.

Stage one. The specimen has been cut out; its blood supply is gone and its enzymes are not, so until it reaches fixative the tissue digests itself. That is autolysis, the window is cold ischemic time, and it has been measured.

Vassilakopoulou and colleagues (2014) built a tissue microarray, many small cores in one block stained in a single run, from 93 breast cancer specimens with recorded time to fixation, and measured 12 phosphoepitopes. An epitope is the patch of a molecule an antibody binds; a phosphoepitope exists only while a phosphate group is attached, which is how a cell carries a signal. Earlier work, they note, saw degradation begin within 30 minutes. In their own array, phospho-AKT, phospho-ERK, phospho-tyrosine and phospho-MET, markers of growth signaling, lost antigenicity, stopped being findable by their antibodies, within one to two hours. Two rose instead, phospho-HSP27, a heat-shock protein, and phospho-S6, a protein of the ribosome, the cell's protein-making machinery, which the authors place in stress-response pathways. A few held steady.

That is human observational evidence. A controlled version exists too. Ndengue and colleagues (2026) cut fifty core-needle microsamples from one invasive ductal carcinoma, a breast cancer, fixed them for half an hour to 144 hours, and delayed fixation in others by half an hour to eight. Hormone receptor and Ki67 staining, Ki67 marking actively dividing cells, held through two hours of cold ischemia, then fell with the delay; across fixation times it held steady to 48 hours and fell significantly only past 72. One patient's tumor, as they say. Haight and colleagues (2024) cut the other way: 941 hysterectomies for endometrial cancer, a cancer of the uterine lining, median cold ischemia 33 minutes, 95 cases at an hour or more, and no significant difference by either in estrogen or progesterone receptor, HER2, a growth-factor receptor, or the mismatch-repair proteins that correct copying errors in DNA. The losses are real; for the markers that decide treatment, small.

One figure you hear elsewhere is citable: the six-to-seventy-two-hour fixation window, with cold ischemia under an hour. Haight's opening sentence gives it as the recommendation of the American Society of Clinical Oncology and the College of American Pathologists, derived from breast cancer data; I found no validation of it for other organs, and their endometrial series is what happened when somebody checked. The same two bodies estimate that up to 20 percent of hormone-receptor results worldwide may be inaccurate, mostly from variation in preanalytic variables, everything that happens to a specimen before the test, in thresholds for calling a result positive, and in interpretation criteria (Hammond and colleagues 2010): a panel's estimate inside a guideline, not an error rate from a trial.

The end of stage one is the pivot. Autolysis is very good at taking a phosphate group off a signaling protein within an hour or two. It is no obvious threat to a worm. A nematode cuticle, nematode meaning roundworm, is neither a membrane nor a protein in solution: Page and Johnstone (2007) describe an extremely flexible and resilient exoskeleton built predominantly of cross-linked collagens plus insoluble proteins called cuticlins. A helminth eggshell, helminth being the word for a parasitic worm, has a measured thickness: Brownell and Nelson (2006), inactivating eggs of Ascaris suum, the pig roundworm, with ultraviolet light, give it as 3 to 4 micrometers, thousandths of a millimeter.

Neither measures what an hour on a cutting bench does to a parasite in human tissue, and I found no study that does. But the asymmetry is structural: the first stage erodes a phosphorylated human epitope within one to two hours, and nothing in it would remove a collagenous exoskeleton or a three-to-four-micrometer shell.

The bench, the jar, and the blade

Start at the bench, because on my reading of the measurements that is where the most tissue is lost. Grossing is where a person opens the specimen, slices it and chooses which slices go into the small plastic cassettes that become paraffin blocks. A cassette, by routine practice rather than a cited figure, holds a slice a few millimeters thick and a couple of centimeters across. Somebody chooses, and that choice, as I read it, is the largest filter in the pipeline; the report says what was examined, not what stayed in the bucket. The Atlas page on sampling error carries the arithmetic and the heterogeneity studies.

The slices have a grain. Kumarapeli and colleagues (2018) chilled mastectomy specimens at minus eighty degrees Celsius for twenty minutes so they could be cut uniformly at four-millimeter intervals and, they add, so small lesions would be fixed and found. Four millimeters is their interval. What to submit, tumor, margins, lymph nodes, a representative piece of normal tissue, is protocol and routine practice rather than a cited figure.

Weaver and colleagues (2011), inside a randomized trial of sentinel lymph node biopsy in breast cancer, removing the first node a tumor drains to, went back to paraffin blocks of nodes already called negative and cut two more widely spaced levels, staining routinely and for cytokeratin, which marks an epithelial cancer cell, one from a lining tissue. Occult metastases, hidden deposits of spread cancer, turned up in 15.9 percent of 3,887 patients, 95 percent confidence interval 14.7 to 17.1, the band where the true figure most plausibly sits. Same block, same patient, two more planes.

The same paper cuts the other way, and both halves matter. Five-year survival was 94.6 percent with occult metastases against 95.8 without: statistically significant, a hazard ratio for death of 1.40, the risk of dying at any point about two-fifths higher, but 1.2 percentage points. The authors concluded their data do not indicate a clinical benefit from examining negative sentinel nodes further. Their answer was not to cut deeper, a decision about human metastases. I found no equivalent study for anything else in the block.

From the other direction, Gerlinger and colleagues (2012) sequenced separate pieces of four primary kidney cancers and their metastases: 63 to 69 percent of all somatic mutations, those a tumor acquires in its own lifetime, were not present in every region. Each piece lacked mutations others carried. Human observational, four tumors: a small series against the assumption that one biopsy stands for a tumor.

Then the jar, where the folklore is thickest. Ten percent formalin is not ten percent formaldehyde. Thavarajah and colleagues (2012), a narrative review whose figures here come from its full text, give it as 3.7 percent formaldehyde in water with about one percent methanol, which the undiluted stock carries at ten percent as a preservative. So formalin is not methanol and water, as a Gemini chat in this show's research files put it. Two more corrections ride along. Formaldehyde is made from methanol by taking hydrogen off it, oxidation, not drying. And methanol is not safe to touch: Oguz and colleagues (2019) report one woman of 58 poisoned through the skin by a methanol-soaked bandage.

What is in the bottle is mostly not formaldehyde. In water each molecule takes up a water and becomes methylene glycol; shedding that water frees the reactive form, a chemical dehydration, not the alcohol dehydration of stage five, though the review adds that the methanol gives a brief alcohol-fixation phase. Shiurba and colleagues (1998) put the balance far toward methylene glycol, penetration rapid and binding slow: the penetration-fixation paradox. Both forms fix, Thavarajah states; the shifting equilibrium sets the rate. Initial cross-linking finishes 24 to 48 hours after penetration, stable methylene bridges, one-carbon links between molecules, may take about thirty days, and the first phase is reversible, the basis of antigen retrieval, the heating step that lets an antibody find a masked protein again.

The third correction is the mechanism. Every account gives it in one sentence: formaldehyde makes a methylene bridge between two lysines. Metz and colleagues (2004), in the test tube, treated model peptides, short chains of amino acids, and identified the products by mass spectrometry, which weighs molecules. They found several kinds of attachment on lysine, arginine, cysteine and histidine, methylene bridges among them, and reported that formaldehyde did not generate intermolecular cross-links between two primary amino groups, the free amino ends lysine carries: exactly what a lysine-to-lysine bridge between two molecules would be. Tayri-Wilk and colleagues (2020) wrote that identifying the cross-links that way had so far been unsuccessful, and showed them to be two formaldehyde-altered amino acids joined together. Metz in the test tube, Tayri-Wilk in fixed human cells grown in culture: the central step of routine pathology was still being worked out in 2020.

The finding that wounds the folk belief comes next. Laboratories worry about overfixation. De Marzo and colleagues (2002) tested it: prostate tissue processed the same day or fixed one, two, three or eight days, then set into two tissue microarrays, 564 cores, so every sample met identical staining. Strong staining for p27, a protein that restrains cell division, rose significantly from same-day processing to one day or more, p below 0.0001, and at eight days beat same-day. For this protein, they concluded, overfixation is much less of a problem than underfixation. One protein, one organ, a direct test against the belief. Not the last word: Ndengue's single tumor lost hormone-receptor and Ki67 staining past 72 hours. Both results concern human proteins; neither touches a worm.

So does the jar destroy the evidence? Three findings say no, and a newer one says the answer is not free. Baszler and colleagues (1999) tested fresh and formalin-fixed tissue from 61 bovine fetuses, cattle, not people, for Neospora caninum, a single-celled parasite that causes abortion in cows. On fixed, paraffin-embedded tissue the polymerase chain reaction, PCR, which copies one chosen stretch of DNA until it can be detected, found it in 13 of 13 truly positive fetuses and one of sixteen true negatives; on fresh or frozen tissue, in 10 of 13 positives and none of 11 negatives. The fixed specimen caught more of the true infections, and the authors offered a presumptive reason: in a histologic specimen, tissue processed onto slides, you choose the lesion under the microscope first.

In humans, Safaei and colleagues (2002) tested paraffin-embedded skin biopsies for Leishmania, a single-celled parasite that lives inside the host's own cells, amplifying DNA from its single large mitochondrion, the compartment where a cell makes its energy. PCR was positive in all 33 where microscopy had seen the parasite, and in 24 of 29 where microscopy found no amastigote, that small intracellular form, though the clinical diagnosis was cutaneous leishmaniasis, the skin form. Sensitivity 92 percent, the share of true infections caught; specificity 100, the share of uninfected cleared. The DNA was in the block all along; whether the sections missed the parasite by plane or the reader by eye, the study cannot say.

The counterweight is from this year. Filippidis and colleagues (2026) pooled 28 studies of panfungal PCR, aimed at DNA every fungus shares: 75.4 percent sensitive on fixed, paraffin-embedded tissue, 18 studies and 852 samples, against 86.5 percent on unfixed tissue, 13 studies and 207 samples. Different studies, so not a head-to-head, and fungi. Lowry and colleagues (2026), validating a PCR for Haycocknema perplexum, a roundworm that invades human muscle, on 22 samples from eight patients, found fresh-frozen samples more sensitive than fixed ones. Fixation does not dissolve the organism. It does cost the DNA test something.

Third, a single case, and here the episode corrects itself. Nagayasu and colleagues (2026) describe a woman of 79 whose enlarging right upper lobe nodule was partly resected; the sections held fragments of a filarial worm, one of the thread-like roundworms, compatible with Dirofilaria immitis, the dog heartworm: thick multilayered cuticle, abundant muscle, prominent internal ridges. Easy to tell as morphology triumphing over fixation. The authors tell it the other way: unusual ridge-like elevations on the outer cuticle made it hard to separate from a related species, Dirofilaria ursi, by morphology alone, and the species call came from a mitochondrial 12S rRNA sequence, part of the gene for the mitochondrion's own ribosome, amplified out of the block. Structure survived, DNA survived, structure alone was not enough.

The step that does destroy evidence is the acid. Decalcification softens bone so a blade can cut it, with a strong acid or a chelating agent, a molecule that grips calcium and carries it off. Miquelestorena-Standley and colleagues (2020) put 35 human samples through hydrochloric acid, formic acid and EDTA, the usual chelating agent: hydrochloric and long formic runs produced false negatives on immunohistochemistry, staining with antibodies raised against one named molecule, and on molecular analysis; EDTA and short formic, under five cycles of six hours, left antigens and nucleic acids usable. Diamantino and colleagues (2024), in dogs' breastbone marrow, found the fixative made no difference to morphology but the decalcifier did, four raters ranking EDTA over formic over hydrochloric; DNA amplified from 29 of 36 EDTA-treated samples, 81 percent, against 2 of 72 acid-treated, 3 percent.

Not hypothetical. Wilkins and Clark (2009) note that a bone marrow trephine, a core of bone with the marrow inside it, must be decalcified or plastic-embedded, and name that as a challenge for immunohistochemistry, for in situ hybridization, which lays a labeled DNA probe straight onto the slide, and for PCR, in a specimen that can hold Leishmania amastigotes; challenges, they add, that careful technique can address. What no study I found has scored is whether an amastigote or an eggshell comes through acid recognizably.

After the jar comes the wax, because water and paraffin will not mix: water is replaced stepwise by alcohol, alcohol by xylene, a solvent paraffin dissolves in, xylene by molten paraffin. Tran and colleagues (2015) prospectively measured 34 kidneys removed for tumors: total shrinkage 11.4 percent, p below 0.01, enough to move two tumors down a size stage on processing alone. So published egg and larval dimensions, measured fresh, will not match a paraffin section; that inference is mine. One detail will not be said here because its citation was invented: shrinkage halos around worms in fixed tissue, attributed by a Gemini chat in this show's research files to Guarner and Brandt (2011), a review of fungal histopathology whose abstract says nothing of such halos.

The blade comes next. Routine sections run three to five micrometers, practice rather than a cited figure; Salehi and colleagues (2016) cut four to five. Thinner than a cell, so the slide is one arbitrary plane, not a picture.

The plane carries its own damage: crush from toothed forceps, coagulated protein from cautery, tangential cuts at a glancing angle that hide the true shape, scratch lines from a nicked knife, chatter from a loose knife or block, fungi and fibers picked up at the flotation bath. Bindhu and colleagues (2013), a narrative review, collect them: a catalog, not a measurement.

Two items on it cut both ways. Floaters are fragments of one case carried into another at grossing, in processing, or on the flotation bath: material does reach slides that was never in the patient, the first objection anyone will raise against a positive parasite finding, and correctly. The second is acid formalin hematin, a dark brown pigment unbuffered formalin's formic acid makes from heme, the iron-bearing part of hemoglobin, which the same review says may simulate microorganisms. Polarized light identifies it, neutral buffered formalin prevents it, and any photograph of organisms in tissue must exclude it first.

Which leaves the gap this cannot close. I found no study that isolates whether xylene or graded alcohol degrades a helminth cuticle or an eggshell. The measured evidence is narrower: the chemistry hardens and cross-links, acid is the one reagent documented to blind the later tests, fixation trims what a DNA test can find, and the losses actually counted fall at the choices: the endoscope, the bench, the blade and the smear.

How many pieces, and the specimens that never become a block

Everything so far assumed a block, and how many pieces go into it is measured too. Lebwohl and colleagues (2011) counted the pieces in duodenal biopsies, the duodenum being the first stretch of the small intestine, bitten off with forceps through a flexible camera, from 132,352 American patients without known celiac disease, an immune reaction to gluten that flattens the gut lining, all sent to one laboratory operating in 43 states. A guideline proposed in 2006 asked for four or more pieces. Four or more went in for 35 percent, 45,995 patients; still 37 percent in 2009, and 39.5 percent even where the stated reason was malabsorption, poor absorption of food, or suspected celiac disease.

And the pieces mattered: a new celiac diagnosis in 1.8 percent of biopsies with four or more pieces against 0.7 percent with fewer, p below 0.0001, in the authors' words a small absolute increase but a doubling of the diagnosis rate. Their limits: retrospective, no clinical follow-up, and the guideline published during the study, possibly confounding it. An association, in one disease. My inference: a forceps bite of gut lining is the specimen most likely to meet an intestinal parasite, and in 65 percent of these cases, by subtraction, fewer pieces went in than asked. One 2026 case fits: Asdaghi Jahromi and colleagues found larvae of Strongyloides, a gut roundworm, in the crypts of a duodenal biopsy after serial stool examinations were negative, without molecular confirmation, a limit they name.

Fine-needle aspiration draws loose cells through a thin needle and smears them on glass for cytology, the reading of loose cells rather than intact tissue; the architecture, the arrangement a section keeps, is lost. Sadiku and colleagues (2026), in Kosovo, reviewed fourteen ultrasound-guided aspirates of cystic liver lesions in a series on hydatid disease, the cyst the larval tapeworm Echinococcus granulosus forms in an organ. The smears were searched for protoscolices, the tiny heads of future tapeworms budding inside the cyst, for its layered wall, and for hooklets, the hooks on those heads. Seven, 50 percent, showed diagnostic features. Seven yielded only nonspecific inflammation, dead-tissue debris, or histiocytes, the scavenger cells that clear debris.

The abstract does not say how, or whether, those seven were confirmed as hydatid; if they were, the loss was sampling, not chemistry. In the authors' phrase, half of aspirates may be nondiagnostic.

Whether somebody scans the whole smear matters too. Pal and colleagues (2018) reviewed five years of cytology in West Bengal, where filariasis, infection with filarial worms, is endemic: 16,738 aspirates and 882 body fluids. Twenty-two held microfilariae, those worms' microscopic larvae: 0.124 percent, every one incidental, from sites including lymph nodes, skin, scrotum, breast, thyroid and pleural fluid, the fluid around the lungs. Two accompanied malignant lesions, which says nothing about cause. Eosinophilia, a raised count of eosinophils, the white cells that rise in allergy and worm infection, was present in 15, 68.18 percent. Retrospective, endemic, incidental.

Meena and colleagues (2026) report three patients in New Delhi, July 2023 to June 2025, each with a single painless swelling of 1.5 to 3 centimeters at the epitrochlear lymph node, above the elbow on the inner arm. Aspiration showed sheathed microfilariae, each in a thin envelope, in all three, and blood smears were negative in all three. All resolved within four weeks on diethylcarbamazine, a drug against filarial worms; the authors note the lump can be mistaken for an infection or a tumor. The worm was in the lump, not in the blood drawn.

Between smear and section sits the cell block: an aspirate or fluid spun into a pellet, then fixed and embedded in wax so it can be cut like tissue. Guerra and colleagues (2019) compared preparations of lymph-node aspirates from dogs, an animal study, for visceral leishmaniasis, the form that settles in spleen, liver and marrow. Smear cytology was 34.00 percent sensitive; cell-block immunocytochemistry, antibody staining of the spun-down cells, 70.0 percent; both together, 72.0. Three cautions: the same abstract calls the smear “the most accurate morphological diagnostic method (45.0%)” without reconciling that with 34.00; it states a detection limit in “amastigotes/mm”, a unit I cannot interpret; and it gives no number of dogs.

In people, the evidence is one case. Singh and colleagues (2024) diagnosed isolated pulmonary cysticercosis, the larval cyst of the pork tapeworm, here in the lung alone, on pleural fluid cytology and a cell block, and argue that in a cell block morphology and architecture are “as clearly discernable as in histopathology.” Four of the five authors work at a commercial laboratory chain; no product is evaluated.

An effusion is fluid collecting where it should not, here around the lung. Bixby and colleagues (2024) open by stating that pleural fluid cytology detects malignancy with an overall sensitivity of 50 percent, depending on tumor load, fluid volume and the reader's experience, and compromised by scarce tumor cells or inflammation. That is their framing of the literature, not their measurement, ahead of their own test, which reads chemical tags on DNA floating free in the fluid: an area under the curve of 0.912, where 0.5 is chance and 1 perfect separation. Two of the nineteen authors list a commercial affiliation, Precision Epigenomics. My inference: scarce material and obscured morphology are when parasite material in a fluid would be missed too.

Cerebrospinal fluid bathes the brain and spinal cord and is drawn through a needle in the lower back, a lumbar puncture or spinal tap. Angiostrongylus cantonensis, the rat lungworm, carried by snails, causes eosinophilic meningitis, inflammation of the brain's lining with eosinophils crowding that fluid. Lin and colleagues (2003), in Chinese with an English abstract, investigated an outbreak in eight pupils in Changle City. Conventional examination of their spinal fluid found two third-stage larvae, the infective stage, in one patient. Eosinophils averaged 68 percent of the fluid's cells, range 47 to 83, in all eight: the host's reaction in every child, the worm in one.

Qvarnstrom and colleagues (2015), at the Centers for Disease Control and Prevention, one author listing Life Source Biomedical LLC, ran real-time PCR, which measures copied DNA as it accumulates, on 49 spinal fluid specimens from 33 patients with eosinophilic meningitis: Angiostrongylus DNA in 32 specimens from 22 patients, and four patients alternated between positive and negative on later taps, so the DNA level may fluctuate. Centrifugation, spinning a fluid so its cells settle, decides what is seen. Mumba Ngoyi and colleagues (2013) examined 94 spinal fluid samples from people with sleeping sickness in the Democratic Republic of the Congo: modified single centrifugation found trypanosomes, the single-celled parasites behind it, in 85 patients, double centrifugation in 46, p below 0.0001.

Marrow is two specimens that see different things: the aspirate, liquid marrow drawn and smeared, and the trephine core. Gupta and colleagues (2008), in New Delhi, reviewed 121 patients with normal immunity and fever of unknown origin, a prolonged fever with no cause found, who had both. The aspirate smear was diagnostic in 16.5 percent, finding leishmaniasis or pure red cell aplasia, marrow no longer making red cells; the trephine, in 76 percent. Granulomas, the walled-off inflamed nodules part one described, appeared in 70 percent of trephines and 1.6 percent of aspirates. The smear is where the amastigotes were recognized; the core, which acid treats, holds the architecture.

Cruz and colleagues (2006) followed 25 children in Spain with visceral leishmaniasis. After treatment with the drug liposomal amphotericin B, clinical improvement went with marrow microscopy and culture turning negative. Yet nested PCR of the marrow aspirate, a two-round reaction that pulls out very small amounts of DNA, showed that sterile cure, no parasite left at all, was not achieved in eight, two of whom relapsed within 10 to 20 weeks. A negative smear means nothing was seen in what was examined. The Atlas page on Leishmania carries these comparisons.

Blood is next. Okell and colleagues (2009) pooled, in a meta-analysis, a study that combines other studies' results, 72 paired measurements from community surveys where the malaria parasite Plasmodium falciparum is endemic, each made by microscopy and PCR. Microscopy found on average 50.8 percent of the prevalence, the share infected, that PCR found, 95 percent confidence interval 45.2 to 57.1; for gametocytes, the sexual forms a mosquito picks up, 8.7 percent. It caught 74.5 percent where PCR prevalence exceeded 75 percent, and 12.0 percent where it was under 10.

Dreyer and colleagues (1996), in Recife, Brazil, found that microfilariae of the filarial worm Wuchereria bancrofti rise in the blood at night and fall by day. Films of 20 and 60 microliters, millionths of a liter, did not reliably detect carriers below 100 or 60 microfilariae per milliliter. Daytime filtration of venous blood found every carrier whose night film had shown more than one; the abstract prints the filtered volume as “1 micromilligram”, not a unit of volume, so I leave it unconverted.

Dumas and colleagues (2020), in Lyon, took 584 samples tested for malaria and asked how often the Sysmex XN-10, the analyzer behind an ordinary blood count, flagged parasitized red cells. Sensitivity 7.8 percent, specificity 100, positive predictive value 100, negative 87.7, a predictive value being the share of flagged samples truly positive, or of unflagged ones truly negative. My inference: most blood counts never become a slide, and the analyzer almost never raises the question itself. Community surveys and a hospital laboratory, not cancer patients.

On the cancer ward, Smith and colleagues (2022) describe a woman of 32 from Guatemala with Philadelphia chromosome-positive B-cell acute lymphoblastic leukemia, a cancer of immature B cells, the white cells that make antibodies, with a rearranged chromosome, treated with a targeted drug plus intensive chemotherapy. In respiratory failure, starved of oxygen, her bronchoalveolar lavage, saline washed into the lung and recovered, showed numerous larvae consistent with Strongyloides hyperinfection. This roundworm can lie dormant in the gut for decades; with suppressed immunity, the authors write, it can cause autoinfection, completing its cycle inside the same person, and progress to hyperinfection, that cycle out of control. She had eosinophilia, and her Strongyloides antibody test was negative. In the authors' words, cytology evaluation is an essential factor for diagnosis. The Atlas page on Strongyloides carries the hyperinfection story.

Grech and colleagues (2000) describe a child in Malta on maintenance treatment for acute lymphoblastic leukemia who developed visceral leishmaniasis. Repeated bone marrow aspirates showed no amastigotes; an immunofluorescence antibody test, a blood test read with a glowing dye, was positive. Two case reports, the same kind of patient, opposite failures: no rate, but neither test closed the question alone.

The last specimen is the body itself, and it needs a correction. A Gemini chat in this show's research files said endoparasites, the ones living inside the body, cannot survive the host's death and die with it. The error is overgeneralization. The same chat's gentler version, that parasites inside a person do not eat their way out when the person dies, is half right: this research found no worm living inside a human documented boring out of a body after death. The half that fails is that they all die with you. Trichinella is a roundworm whose larvae coil in muscle inside a nurse cell, a muscle cell the larva remodels into a shelter. Rossi and colleagues (2019) left infected fox and raccoon dog carcasses beneath or above mountain snow, and every 14 days fed recovered larvae to mice to see how well they still multiply, a reproductive capacity index. Beneath the snow it fell from 23 to 25 at day 0 to 12 to 18 at day 112; above it, from 22 to 27 to 0.0; p below 0.01. Animal carcasses, not people. Pozio (2000) gives the mechanism: the larvae's anaerobic metabolism, running without oxygen, lets them survive in extremely decayed meat. Encysted Trichinella larvae, at least, do not die with the host.

A caveat runs the other way, for any molecular finding at autopsy. Javan and colleagues (2018, in print 2019) review the thanatomicrobiome, the body's microbes after death: they vary by organ, time and temperature, and the gut where small intestine meets large carries the largest burden, spreading to liver and spleen, then heart and brain, depending on the cause of death. My inference: bacterial DNA from an autopsy organ may have arrived after death, so an autopsy finding is only as good as the lesion it came from. The series part five closes on took its DNA from inside granulomas, walls the living body built, and one organism was a heartworm, no part of the gut flora that spreads after death.

Every loss counted here is volume, timing, spin or specimen; none is a fixative dissolving a worm.

The stain, the reader, and the specimen that reached them

The stain comes next. Two dyes do most of the work in every pathology laboratory: hematoxylin and eosin, H and E, nuclei blue-purple and the rest pink. The pink is eosin, which binds protein. The blue is hemalum, hematein mixed with an aluminum salt, and the textbook calls that aluminum a mordant, a metal ion gripping the dye at one end and the chromatin, the DNA and protein inside a nucleus, at the other. Kiernan (2018) tested it and the theory did not survive: aluminum first and hematein after stained more weakly than hemalum itself; hematein first and aluminum after gave nothing.

What stains is a bulky red dye-metal complex, cationic, meaning positively charged, formed in a solution carrying sixteen aluminum ions for every hematein molecule, and too large, on his reading, to slip between DNA's base pairs. Its target looks like DNA rather than protein: nucleases, enzymes that digest nucleic acids, abolished the staining, while methylation, a treatment that caps the tissue's acid groups and completely blocked the simple dye thionine, did not. Bench chemistry, one investigator, but on the routine stain of nearly every slide, and the textbook mordant account did not survive it. The Atlas page on hematoxylin and eosin carries the rest.

Which matters for one reason. H and E is tuned to nuclei and protein; it is not an organism stain. An egg on that slide is a shape, pink and purple like everything around it, legible only to a reader who knows it. Grogan and colleagues (2000) put the repertoire at twenty to twenty-five special stains kept on hand, acid-fast for the tuberculosis bacterium, trichrome for scarring. A special stain is a question, and somebody has to ask it.

Asking is no guarantee. Özmen and colleagues (2013) ran five methods on 100 bronchoalveolar lavage and bronchial washing specimens for Pneumocystis jirovecii, the fungus that causes pneumonia in the immunosuppressed. Methenamine silver, toluidine blue O and Wright-Giemsa, three stains used to find it, showed it in no sample. A fluorescent antibody found cysts in 13 patients and nested PCR found DNA in 16, only 8 by both; against the antibody test as their standard, PCR was 61.5 percent sensitive, and the authors concluded the two do not run parallel. Bronchoscopy patients, most not immunosuppressed, none with HIV, the human immunodeficiency virus, and the subject was colonization, the organism present without disease, not pneumonia. Three stains that found it in nobody, and two better tests that disagreed.

A Gemini chat in this show's research files offered a tidier version, and it was false: a 2000 study of 110 brain biopsies in which molecular testing beat histology, 97.5 percent against 75. No paper with those numbers is indexed in PubMed. The nearest real study, Cingolani and colleagues (1996), tested spinal fluid, not brain, in 88 patients with HIV: nested PCR for Toxoplasma, a single-celled parasite that can infect the brain, was 33.3 percent sensitive and 100 percent specific. Brain biopsy, in the same group's 136-patient series, was 93 percent sensitive across all causes of localized brain lesions, and done only after three weeks of failed anti-Toxoplasma treatment (Antinori and colleagues 1997): no like-for-like comparison. The fabricated number ran in exactly the direction this part must watch.

The same material read two ways gives two answers. Oztek and colleagues (1997) reviewed sputum, bronchial washings and bronchial brushings from 131 patients with suspected complicated lung hydatid disease, 111 proven on histopathology. Hooklets turned up in 26 patients on Papanicolaou slides, the stain used for cervical and respiratory cytology; in 37 on trichrome; in 50 under dark-field microscopy, which changes how light crosses the slide rather than staining anything.

The reader is the best step in this chain and the weakest. The best: Brunyé and colleagues (2021) tracked twelve pathologists' eyes, ten of them in training, over 48 digitized biopsies of the skin's pigment cells shown for half a second each; in that time their eyes went to the critical abnormality, and they sorted the images for invasive melanoma with high sensitivity and specificity: a mechanism, not an error rate. The weakest: Elmore and colleagues (2015) had 115 pathologists in eight American states independently read 240 breast biopsies in test sets, one slide per case, told only the woman's age and the type of biopsy: 6,900 diagnoses. Agreement with an expert panel's reference was 75.3 percent, 95 percent confidence interval 73.4 to 77.0: invasive carcinoma 96 percent, but atypia, cells that look abnormal but not malignant, 48 percent, with 35 percent underinterpreted. The three-member panel was itself unanimous only 75 percent of the time. Agreement was lower on denser breasts, 73 percent against 77, and for readers with smaller caseloads or practices.

The sampling half has been audited too. Novis, Gephardt and Zarbo (1996) collected 18,532 frozen section diagnoses, readings made on tissue frozen and cut while the patient is still on the table, from 233 small hospitals, all but one in North America. Of those, 859, 4.6 percent, were deferred to the permanent section, the slower wax-process slide; of the ones answered, 316, 1.8 percent, disagreed with the wax slide. The most common cause, in the authors' words, was underdiagnosis of neoplasia, tumor growth, “usually due to block- or tissue-sampling errors.” From 1994, at hospitals of three hundred beds or fewer.

Recognizing non-human tissue is its own problem, and the nearest measurement concerns fungi. Salehi and colleagues (2016) ran PCR for fungal ribosomal DNA on 102 formalin-fixed specimens that already carried a histopathology diagnosis. Fusarium, one mold, came out of five read as aspergillosis, infection with another, Aspergillus; and Aspergillus, Scedosporium or Syncephalastrum out of others read as mucormycosis, a third kind of mold infection. The histopathological features of molds, they conclude, may be easily confused in tissue sections.

The hardest sentence belongs to the Medellín case from part one. Muehlenbachs and colleagues (2015) describe nests of monomorphic undifferentiated cells, all alike and of no mature tissue type, in lymph-node and lung biopsies from a man with HIV, behaving like cancer; only their small size suggested a nonhuman origin. A PCR aimed at eukaryotes, every organism whose cells carry a nucleus, identified Hymenolepis nana, the dwarf tapeworm. The clause to read slowly: although the cells were unrecognizable as tapeworm tissue, antibody staining and a labeled DNA probe marked them in place once the laboratory knew what to look for. One patient, and the clearest case that a thing can sit on the slide, before expert eyes, unrecognized.

The pipeline can counterfeit a parasite as easily as hide one. Abreu Velez and colleagues (2009) found unexplained pigment on routine slides from ten patients and ten matched controls in El Bagre, Colombia. No organism was found; iron stains and melanin bleaching failed; the pigment came away only after very strong acid, identifying it as formalin pigment, the deposit acidic formalin makes on meeting blood. Their own sentence is the warning: it shares many properties with hemozoin, which is malaria pigment.

The head-to-heads come next, one group measured three ways. Poggensee and colleagues (2001) examined 228 women in an area of Tanzania endemic for Schistosoma haematobium, the blood fluke whose eggs lodge in the wall of the urinary and genital tract. Crushing the fresh cervical specimen flat under glass, the quantitative compressed biopsy technique, found eggs in 112, 49 percent; histology of a cervical biopsy, the ordinary fixed-and-sectioned route, in 40, 18 percent; a cervical smear in 6, 3 percent. Median egg load was significantly higher where histology got it right than where it missed: light infections get missed.

Correct the misattribution that travels with that number, by the name of the error. The 3-against-49 comparison is a cervical smear against a crushed cervical biopsy, not urine: a specimen-type substitution, real numbers, wrong specimen. Pillay and colleagues (2016), 394 women aged 16 to 23 in rural KwaZulu-Natal: urine microscopy found eggs in 78, 19.7 percent, Papanicolaou smears in 8, 2.0 percent. Kokubo-Tanaka and colleagues (2025), 269 Kenyan schoolchildren: 50.2 percent carried circulating anodic antigen, a molecule the living worm sheds into blood, against 14.1 percent with eggs in urine. And whole-organ, El-Bolkainy and colleagues (1981): eggs in 902 of 1,095 Egyptian bladder cancers removed with the whole bladder, 82.4 percent. The Atlas page on Schistosoma haematobium carries the head-to-heads.

What moves across all of it is how much tissue was examined. Gomez and colleagues (2017) ran a sequencing assay aimed at a DNA stretch shared by all fungi on 233 specimens; among the 116 with only suspected disease, yield was 71.5 percent for open biopsy, 40 of 56, against 50 percent for core-needle, 17 of 34, p of .04. Kim and colleagues (2023) biopsied one lung nodule two ways: Paragonimus westermani eggs came from the cryobiopsy, which freezes tissue onto a probe and pulls a larger piece free, not from the forceps bite, a difference the authors put down to specimen size. A single case, as is Cong and colleagues (2022), where the fluke appeared only on the third needle biopsy.

Then coexistence without causation. Srougi and colleagues (2017) report a young patient from central Africa whose bladder biopsy showed both a poorly differentiated carcinoma, cells that have lost the features that would say what tissue they came from, and Schistosoma eggs. Ahluwalia and colleagues (2003) found two sheathed microfilariae of Wuchereria bancrofti on the aspirate of a Ewing's sarcoma, a bone cancer; Singh and colleagues (2010), the same worm's larvae beside adenocarcinoma cells, a cancer of gland-forming tissue, in chest fluid. Three case reports: a parasite and a tumor can share a slide, which says nothing about how often or about cause.

Rank the losses as I read the measurements; no study puts sampling and chemistry on one scale, so the ranking is mine. Sampling first by a wide margin, what the knife selected and what reached the block. Non-recognition second, a reader asked a different question. Chemistry a distant third: acid decalcification is the one reagent measured to blind the later tests, fixation trims the DNA test's sensitivity, long heating in staining costs eggs, and archive age, which part five measures, wears the DNA down further. Iodine is not on this list; part five takes that claim apart.

Here is the hard limit: none of this gives a rate. I found no study that takes a consecutive series of tumor resections and asks how often expert or molecular re-examination finds parasite material the original report missed. The two nearest numbers do not cover it. Rodriguez and colleagues (2014) ran a parasite PCR on paraffin blocks from presumptive abdominal angiostrongyliasis, bowel infection by a relative of the rat lungworm, where only the tissue reaction was visible: positive in 4 of 20, a 20 percent gain, from an assay itself 55 percent sensitive. And Poggensee's counts put sections at about 36 percent of the crushed-biopsy reference, my arithmetic, in a cervix, not a tumor. Neither may be moved onto one.

The honest half: the documented errors I found run mostly the other way, parasitic lesions read as cancer. Silver and colleagues (1996): a child's solitary brain lesion biopsied as malignant glioma, a cancer of the brain's supporting cells, and cysticercosis on excision. Ermilov and colleagues (2018): a frozen section suggestive of glandular cancer, a resection on that reading, a patient dead on postoperative day two of a clot lodged in the lung's arteries, and paragonimiasis found after death. Then denominators. Hara and colleagues (2022), nationwide Japanese data on 1,699 surgical pleural biopsies: paragonimiasis 0.1 percent. Qin and colleagues (2018): one sparganosis, infection with the tapeworm larva from part one, in 29 consecutive deep brain biopsies. And one case runs in reverse: Wang and colleagues (2022), a herder of 69 whose liver cancer hid on every preoperative imaging test behind a hydatid diagnosis.

The worms that were found

The cases that went right come next; the show owes them too. When a worm is in the tissue and the tissue reaches the glass, it usually gets named. Yıldız and colleagues (2015), in Turkish, reviewed the pathology of 846 children whose appendix was removed for appendicitis from 2008 to 2014. Fourteen had parasitic appendicitis: Enterobius vermicularis, the pinworm, in twelve, and Taenia, the genus of the beef and pork tapeworms, in two. In three children with an inflamed appendix, a parasite was found as well, two pinworms and one Taenia.

The abstract's own figure, parasites in 0.39 percent of patients with appendicitis, matches neither 14 of 846, 1.65 percent, nor 3 of 846, 0.35, so the counts go on air and the percentage does not: a pathology service reading fixed appendices named a roundworm twelve times and a tapeworm twice.

Cytology shows the same for a protozoan, a single-celled parasite. Tawfik and colleagues (2014), in Kansas City, read 335 whole-slide images, entire slides scanned into one digital picture, of Papanicolaou cell blocks against ThinPrep, a liquid-based cervical cytology laid down by machine in a thin layer. Bacterial vaginosis, an overgrowth of the vagina's mixed bacteria, was called on 33 against 36; yeast on 18 against 19; Trichomonas vaginalis in 10 of 11 samples on both. Two preparations against each other, at one center, with no independent reference. My inference: a protozoan the screening is built to report was reported.

A tapeworm larva: Boonyasiri and colleagues (2014) describe nine human cases of sparganosis at one Thai hospital, 2001 to 2012, each diagnosed by histopathology first. DNA from the same fixed, paraffin-embedded tissue then yielded a fragment of cox1, a mitochondrial gene whose sequence differs enough between species to name them: all nine were Spirometra erinaceieuropaei. Nine cases, so no denominator and no sensitivity. But it is the chain working for a worm: fixed, waxed, cut, read, recognized, then named to species out of the block.

And one from this year, from a tumor operation. Beltrán-Rosel and colleagues (2026) describe a woman of 30 from Côte d'Ivoire, living in Spain, whose leiomyoma, a benign tumor of the uterine muscle, was removed during a workup for infertility. The pathologist found schistosome eggs with granulomatous inflammation in the muscle specimen, though her first antibody test and urine examination had been negative; later urine samples confirmed the infection. One patient, and the slide answered a question nobody had asked.

That is the easy direction. The field's current statement of the hard one is a French review, Trecourt and Radobonirina (2025), quoted from its English abstract. True cysts, lined by the parasite's own tissue, are formed in people by tapeworms that cannot finish their life cycle: hydatid disease, alveolar echinococcosis, whose larva infiltrates the liver, and cysticercosis. A pseudocyst is walled off instead by the host's own granulomatous reaction, which roundworms and tapeworms both provoke.

These pathogens are “rarely encountered by pathologists,” the authors write, with “many morphological pitfalls.” Their remedy is referral to an infectious-disease pathologist or a parasitologist used to tissue sections, with morphology and DNA read together where possible. A narrative review, not a measurement, and no rate; what it gives is pathologists saying in print, in 2025, that the hard step is recognizing the thing.

Here is the hard case at its purest: the slide that showed the host and not the parasite. Luo and colleagues (2026), at West China Hospital in Chengdu, report a woman of 38 with a mass on the left side of her liver whose first scan was suspicious for intrahepatic cholangiocarcinoma, a bile duct cancer arising inside the liver; the scan belongs to part six. Her tumor markers were normal. The tissue showed necrotizing granulomatous inflammation, granulomas with dead centers, infiltrated by eosinophils, and Charcot-Leyden crystals, the slender crystals eosinophils leave as they break down. No schistosome eggs were identified.

The diagnosis of focal hepatic schistosomiasis rested on two things outside the slide: a positive blood test for IgG, the long-lasting class of antibody, against Schistosoma japonicum, the Asian blood fluke, and her residence in a region historically endemic for it. The authors call the diagnosis presumptive, rightly. On my reading, an antibody says the body has met the worm, not that the worm made this mass; the organism was never demonstrated. Santoro and colleagues (2026) show the other ending: a brain lesion under a centimeter, suspicious for metastasis, a slide of necrosis and inflammation, and antibody, stool and fungal DNA tests all negative, until mitochondrial DNA from the block named a wildlife tapeworm, Taenia martis. In both, the answer came from a test somebody chose to order, which is part five's subject.

Tissue unlike the worm it is, a second time, in an animal. Crouch and colleagues (2021; the printed issue is 2022), at the Bronx Zoo, report an aged ring-tailed lemur that died of disseminated Hymenolepis nana, a species previously reported to shed it in the feces without disease. The parasites were, in the authors' words, “highly unusual”: variably sized, ellipsoid to snaking, lined by a thin tegument, a tapeworm's living outer covering, and filled with pink fibrous filler and blue-staining round cells. The name came from a PCR on the 18S rRNA gene, a ribosomal gene every cell with a nucleus carries, varied enough to tell species apart; the sequence matched H. nana 100 percent. One animal, and an infection that spread, not a cancer.

Set it beside the Medellín cells: same species, different host. In the man the cells did not look like a tapeworm at all; in the lemur they looked like a parasite but like no particular one. In both, the name came from a sequence. The Atlas page on the dwarf tapeworm carries the animal cases beside the human one.

One measurement decides how much of this the eye can do, and it is burden. Moreira and colleagues (2007) studied 89 dogs in an endemic area of São Paulo State, Brazil, with 26 controls from a non-endemic area: an animal study, of Leishmania, not a worm. The dogs were grouped as symptomatic, oligosymptomatic, showing few signs, or asymptomatic, and a lymph node behind the knee was aspirated and stained. The stained aspirate showed the parasite in 75.61 percent of symptomatic dogs, 32 percent of oligosymptomatic and 39.13 percent of asymptomatic; immunocytochemistry in 92.68, 76 and 100 percent; PCR in 100, 96 and 95.65. Every control was negative on every method.

Three cautions. The groups are defined by how sick the dog looked, not by parasite count, so burden is inferred, and the milder two are out of order, 32 against 39. Enrollment leaned partly on the very smear being measured, which on my reading flatters it. And the perfect specificity rests on 26 controls. What survives is the shape: in sick dogs the eye found three in four, in the others about one in three, while PCR held at 95 percent or above.

That is the Tanzanian cervix again, another host and organism, where light infections were the ones sections missed. It is also why the host's reaction so often reaches the reader first: the eosinophilia beside most of the microfilariae, the eosinophils and crystals in that liver. My inference, not a measurement: when the organism is scarce, what morphology sees is mostly the body's answer to it.

The claim read at the start of the hour said formalin kills and dissolves whatever is in the jar, so a worm inside a tumor is gone before anyone could look. On everything this part has opened, no fixative or processing step dissolves a worm, though long heating in staining costs eggs and fixation trims what a DNA test can find; and even that is an inference, since the split-specimen study has never been run. Mostly, the chain does not destroy. It chooses: how many pieces go into wax, how much fluid is spun, smeared or poured away, which hour the tissue reaches the jar, which plane the blade cuts, which question the reader was asked. How often those choices add up to a parasite missed in a tumor is the rate nobody has.

What a slide can say next depends on the tests somebody orders: the special stain, the antibody, the PCR, the panel. Part five is those tests. Formalin keeps the worm. The choosing is what loses it.

Sources

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