Twelve Chromosomes, and the Reads Nobody Reads

Part five takes the tests somebody has to order once tissue reaches the glass, and the claim that iodine has already destroyed whatever they might find. It opens on a kidney-transplant patient whose lung cavity and adrenal mass were treated as an Aspergillus infection, on the strength of a positive antigen test, until a wedge resection found a lung fluke. Then the antibodies a pathologist chooses, and why a marker reports its epitope rather than the tumor. Then karyotypes against gene panels, and why a wax block cannot be karyotyped; parasite versions of human cancer genes, and why that is not the same as a primer binding; and the reads a sequencer throws away, with the retracted 2020 tumor-microbiome paper as the cautionary tale. Then the stool exam's own history, from a 1948 Army sedimentation method to the cleared panels whose menus we checked, which contain no worm, and the first marketed panel that does. Then the iodine claim taken apart in five corrections and a randomized trial, the dwarf tapeworm and whether American hospitals look, and the study nobody has run.

Show notes

Part four opened on a lung fluke's eggs, sitting on tuberculosis slides somebody had already read. This part opens on the same fluke, on purpose. Paragonimus westermani is a flatworm caught by eating raw freshwater crab, and it most often invades the lungs. Kwon and colleagues (2018), at a Korean university hospital, describe a man of 65 with a kidney transplant behind him who came to clinic coughing blood. The paper calls him immunocompromised: his defenses against infection turned down. A CT scan, computed tomography, X-rays assembled by a computer into slices of the body, showed a cavity in one lung and a mass on the right adrenal gland, the hormone gland that sits on top of the kidney.

Two stories fit that picture, and the authors name both. Lung cancer with a metastasis in the adrenal, meaning cancer spread from where it began and rooted in another organ. Or a fungal infection in the lung with an adrenal adenoma beside it, a benign growth of the gland. A PET-CT — positron emission tomography, which maps where a radioactive tracer collects, laid over the CT slices, and part six's subject — suggested benign disease. Bronchial lavage is saline washed into the airways through a bronchoscope, a flexible camera, and drawn back out. In that fluid the antigen titer for Aspergillus, a mold, was elevated and, in the paper's phrase, showed positive conversion. An antigen is the molecule a test is built to recognize, here one of the mold's; the titer is how much was measured. On that result the case was diagnosed as pulmonary aspergillosis. The authors' own word is misdiagnosed.

He was treated for the mold, the treatment brought side effects, and because of them he had a wedge resection, an operation that takes a small wedge of lung for the pathologist. In it was P. westermani. Then, in the order the paper gives, a careful history: he had eaten raw freshwater crabs three years earlier. A test for serum antibodies to the fluke was positive — an antibody being a protein the immune system makes to fit one foreign molecule, and serology the testing of serum, the clear fluid of the blood, for them. He was given praziquantel, the tablet that kills flukes, whose story is part three's. Three months later the adrenal mass was still there on CT, and a right adrenalectomy, the removal of that gland, confirmed the same fluke. Two operations, two organs.

Hold what the antigen test did. It was asked whether a molecule of Aspergillus was in that fluid, and it answered. The record does not say why it was positive, and I will not guess. As the paper tells it, the fluke’s question was asked only after a piece of lung had come out.

Now the same fluke, and the other question. Kim and colleagues (2011), at Pusan National University, describe a woman of 27 with two months of cough and sputum, what is coughed up from the lungs. Her CT and her FDG-PET, the version of that scan that uses fluorodeoxyglucose, a radioactive sugar, made her look like a patient with a metastatic lung tumor. She was diagnosed with P. westermani infection by an immunoserological examination using ELISA: an antibody test read by a color change, where a fluke antigen is fixed to a plastic well, the patient's serum goes in, and if antibodies stick a later step turns the well colored. After praziquantel, her follow-up X-ray and CT showed, in the paper's words, an obvious improvement. No operation appears in the record. A tablet, because somebody ordered that test.

Two cases cannot stand in for a rate, and the nearest thing to a count is small. Shu and colleagues (2021) went back through eight adults with lung masses diagnosed with paragonimiasis, infection with this fluke, in the oncology department of a hospital in Yunnan, southwestern China, between July 2014 and July 2019. All eight came from areas where the disease circulates and had a confirmed history of eating uncooked freshwater crabs; their illness had run long and their eosinophils were raised, eosinophils being white blood cells that climb in allergy and in worm infections. They had been diagnosed as pneumonia or tuberculosis and treated for years without improvement, and reached oncology because they were coughing blood. One hospital, eight people, five years: a count of who came through one door, not a rate.

The review literature says where the clue sits. Nakamura-Uchiyama and colleagues (2002), writing a Japanese perspective, say that when paragonimiasis is misdiagnosed as tuberculosis or lung cancer, patients suffer a considerable burden of long-term hospitalization and unnecessary examinations and treatments. Their advice is to think of the fluke whenever lung lesions come with eosinophilia and a raised serum IgE, immunoglobulin E, the class of antibody behind allergic reactions. Both clues are blood tests; neither is on the slide. A review states this rather than measures it, and gives no counts.

And the antibody test is not a general test for parasites either. It is built, molecule by molecule, by people who chose what to put in it. Curtis and colleagues (2021), at Washington University in St. Louis, picked five proteins of Paragonimus kellicotti, another lung fluke, made them recombinantly and tested them by Western blot against sera from lung fluke patients in Missouri and the Philippines, from patients with two other fluke infections, and from healthy North American controls. Two separated the patients best, a cysteine protease and a myoglobin, and a third marked only infections with egg-producing adults. A laboratory study of candidate antigens, in the authors' word promising, with no rates in its abstract. The paper's own opening puts paragonimiasis at 23 million people, mainly in Asia, “often confused with lung cancer or tuberculosis” — their figure, with no source stated in the abstract.

Here is the thesis, said once. Every test in a cancer workup is a question about something named, and a negative result is an answer to that question, not to any other. The Aspergillus antigen answered about a mold; the ELISA answered about one fluke, because somebody chose to ask it. Part four followed the specimen from the body to the glass. This part is what gets asked of it once it is there: the antibodies, the chromosomes, the genes, the stretches of sequence a machine reads and the software sets aside, the stool. And the claim that iodine erased the answer before anyone asked. The claim, in full: “If a parasite were in a tumor, modern testing would find it — the panels read the DNA. They come back clean because the iodine on every specimen has already killed the parasite.”

And the honest frame, up front. No study measures how often a negative cancer workup hides a parasite; the research behind this series looked and found none, and what exists is case reports and a few small denominators, which part four gave. The cases in this open are written up because something in them was unusual — Kim's team say theirs was the first lung fluke on that scan to pass for cancer that had spread, where every earlier one passed for a cancer that began in the lung. And they run the same direction: a parasite taken for something else, named in the end because somebody asked its question, or because a surgeon removed tissue in which it could be confirmed.

The antibodies somebody chooses

When the routine stain cannot say what a tumor is, the next questions are antibodies. Part one defined it, in one clause: an antibody raised against one named human molecule, made visible with a color, run beside a positive and a negative control. What makes it a test is a rule. Fitzgibbons and colleagues (2014) wrote it for the College of American Pathologists: every immunohistochemical test must be validated before it touches patient specimens, on cases whose expression, fixative and processing suit its intended use — enough, in the guideline's words, to ensure it “accurately measures the analyte of interest”.

Goldsmith and colleagues (2024) updated it, adding guidance for cytology specimens, loose cells rather than solid tissue, and for verifying assays the Food and Drug Administration has approved or cleared. Now my inference, not the guideline’s: validation is done against human tissues of known expression, and nothing in either record contemplates a target that is not human. One of the update’s fifteen authors lists a company, Array Science, and one a laboratory consultancy; the rest, hospitals, universities, pathology practices and professional bodies. PubMed records carry affiliations, not disclosures, so affiliations are all I checked.

Nobody orders one antibody. A workup is an algorithm, and each step spends tissue. Selves and colleagues (2018), at French university hospitals and cancer centers, lay out the sequence for a carcinoma, a cancer of the lining tissue, that has spread from somewhere nobody can yet see. The first antibodies ask the broadest questions: carcinoma, or lymphoma, a cancer of white blood cells, or melanoma, the cancer of pigment cells. Then two cytokeratins sort the carcinomas: whether a tumor carries CK7, CK20, both or neither is the first fork toward an organ, and organ-specific antibodies follow. The review's warning is the line this episode needs: with small samples, panels must preserve tissue for molecular tests. Each antibody takes its own slice, and a needle biopsy holds only so many.

What is one answer worth? A marker reports its epitope, the small patch of protein the antibody grips, not the tumor, and one paper argued for reversing practice on exactly that. Bishop and colleagues (2011) stained 470 archival tumors: 81 lung squamous cell carcinomas, cancers of the flat, scale-like lining cells; 237 adenocarcinomas; and 152 large cell lymphomas. The squamous marker most often recommended, p63, was positive in 100 percent of the squamous carcinomas, 31 percent of the adenocarcinomas and 54 percent of the lymphomas: sensitivity 100 percent, specificity 60. The challenger, p40, an antibody against one shorter form of the same protein, was positive in 100 percent, 3 percent and none: specificity 98. The authors recommended p40 in place of p63 for routine use. More than half of a series of blood cancers stained with the marker for squamous carcinoma. A stain reports reactivity; what failed was reading reactivity as identity.

It runs across organs, in both directions. SOX10 is a transcription factor in cells descended from the embryo's neural crest, and Cimino-Mathews and colleagues (2012) note it is used clinically mainly to support a diagnosis of melanoma. They found it in 40 of 100 invasive breast carcinomas, concentrated in the basal-like, triple-negative and metaplastic types, breast cancers without hormone receptors or the HER2 growth receptor. Breast cancer, they concluded, must be considered for any spread tumor positive for SOX10 and for S100, its usual partner. The other direction: Palmirotta and colleagues (2020) gathered 295 patients with pulmonary enteric adenocarcinoma, a rare lung cancer expressing CDX2, CK20 or MUC2, the gut's markers, often indistinguishable from colorectal cancer spread to the lung. A lung cancer answered yes to the gut's questions.

Then the answers that are no. Lao and colleagues (2014), in Chinese, in a record with no DOI, reviewed 22 sarcomatoid mesotheliomas, cancers of the thin lining of the chest or abdomen built of spindle-shaped cells. All 22 stained for pan-cytokeratin AE1/AE3, which asks whether a tumor is epithelial at all; the mesothelioma markers were patchy; and all 22 were negative for TTF-1, napsin A and surfactant protein A, markers of lung origin, for p63, and for CD34, a blood-vessel marker. Five no answers on a real cancer, diagnosed by excluding look-alikes. A no answers one question, and a yes can miss: Li and colleagues (2014) pooled 10 studies in a meta-analysis, and for telling lung adenocarcinoma from squamous carcinoma, TTF-1 plus napsin A had a pooled sensitivity of 0.76, 95 percent confidence interval 0.69 to 0.83, against a pooled specificity of 1.00, interval 0.92 to 1.00. Roughly a quarter missed, by my subtraction.

When the panel does not answer, the cancer is named for the absence. Carcinoma of unknown primary is a cancer found where it has spread, whose starting organ is never located. Pavlidis and Pentheroudakis (2012) put it at 3 to 5 percent of all malignant epithelial tumors; 80 percent are adenocarcinomas, and 80 percent fall in the unfavorable subset, given empirical chemotherapy chosen without knowing the organ, with generally poor responses and survival. Rassy and Pavlidis (2020), a narrative review, gives the ceiling on antibodies: a single tissue of origin predicted in 10.8 to 51 percent, while molecular classifiers identified a primary in 61 to 89 percent and have not been validated in randomized controlled trials. Pavlidis wrote both reviews. Antibodies still come first, being cheap, fast and broadly available.

The classifiers are where this turns, because they are closed sets, each a program trained on tumors of known origin that assigns a new tumor to whichever known type it most resembles. Some read DNA methylation, in a pattern characteristic of the tissue a tumor came from. Moran and colleagues (2016) trained the EPICUP classifier on 2,790 tumors of 38 types and validated it on an independent 7,691 at 99.6 percent specificity. Applied retrospectively to 216 cancers of unknown primary, it predicted a primary in 188, 87 percent; for the other 28 it named nothing. Patients then given therapy for the predicted type had better overall survival than those given empiric therapy; the hazard ratio was 3.24, interval 1.42 to 7.38, p of 0.0051 — observational, not randomized, with funding that included the drug company Ferrer. My point, not the paper's: when EPICUP answers, it answers with one of 38 human classes, and its 99.6 percent measures how well it tells those apart, not whether the tissue holds anything else. Capper and colleagues (2018) built one for the central nervous system out of 82 tumor classes: in routine use it changed the diagnosis in up to 12 percent of prospective cases, and for another 12 percent, 127 of 1,104, it matched no class at all. A classifier can report none of these; it can never report what else is there. The same shape in leukemia: Steinicke and colleagues (2025) built 38 methylation classes from 2,540 samples. Thirty-eight classes, all human.

Then the trial that made a gene panel the first question. Krämer and colleagues (2024) ran CUPISCO, a randomized phase 2 trial, at 159 sites in 34 countries outside the United States, in unfavorable non-squamous cancer of unknown primary. Every patient had comprehensive genomic profiling, a gene panel reading a long pre-chosen list of named human cancer genes. Of 1,505 screened, 436 whose disease was controlled by three cycles of platinum chemotherapy were randomized 3 to 1 to therapy matched to the panel's findings or to more chemotherapy. Median progression-free survival, the time until the cancer grows or the patient dies, was 6.1 months against 4.4: hazard ratio 0.72, 95 percent confidence interval 0.56 to 0.92, p = 0.0079 — a surrogate endpoint for living longer, investigator-assessed in an open-label trial, follow-up ongoing. Roche funded it; five authors list Roche and one Foundation Medicine, a genomic-profiling company. The authors recommend profiling at diagnosis. So the first question asked of a cancer with no known origin is now a list of human genes.

So the gap, said plainly. I could not find a study counting how often a uniformly negative antibody panel is followed by an organism stain, one of the cheap chemical stains that color fungi, bacteria or parasites, or a prospective series adding organism stains and broad-range PCR to undifferentiated tumors, whose cells have lost the look of any tissue. PCR, the polymerase chain reaction, copies one stretch of DNA over and over; broad-range means a stretch shared across whole kingdoms of life, so it can return an organism nobody named. The Atlas page on how cancer is tested carries the marker list and the panel algorithm.

Chromosomes, genes, and the reads nobody reads

Two more tests sit at the end of the chain part four walked, and somebody has to order them too. One counts chromosomes, the other reads genes, and they cannot be run on the same material. A karyotype is a photograph of one cell's chromosomes, seen one at a time only while the cell divides, at the stage called metaphase, so it needs living cells arrested mid-division. A paraffin block is dead tissue cross-linked by formaldehyde and set in wax, and nothing in it divides: a wax block cannot be karyotyped at all. One recent paper says as much in passing, a 2025 Modern Pathology evaluation putting it that chromosome banding has low resolution and relies on viable cell culture (Verhasselt and colleagues 2025). The rest I attribute to cytogenetics teaching and laboratory standards.

The published evidence arrives sideways, in what people had to do to get a karyotype at all. Kaewkong and colleagues (2012) published the first fully named karyotype of Opisthorchis viverrini, the Southeast Asian liver fluke: hamsters infected, adult worms recovered, testes dissected, because that is where cells reliably divide. Six chromosomes in the cells making sperm, where the pairs have already separated, and twelve in a diploid cell caught in mitosis, also from the testes. Twelve is the number that matters.

One fluke-associated human cancer that I could find has been karyotyped, and how it happened is the argument. Sripa and colleagues (2005) took biopsy and bile from a 65-year-old Thai woman with cholangiocarcinoma, cancer of the bile ducts; her bile held malignant cells together with O. viverrini eggs. The cell line was established four months after the primary culture, and only then could the cytogenetics be done: aneuploid karyotypes, with a modal count of 78 against a human forty-six. Four months of culture to manufacture the dividing cells the test requires.

Then the same split inside one laboratory. Haley and colleagues (2021) validated a multiplexed gene-fusion panel, which looks for two genes abnormally joined, then ran it on a consecutive series of 214 — 73 leukemia and lymphoma specimens, 141 formalin-fixed, paraffin-embedded solid tumors — with conventional karyotyping among the parallel comparators. The paper does not say which specimens got which comparator, but karyotyping is among them and only the solid tumors are described as fixed, so the split is an inference from the specimen types, not a rule anybody wrote down.

That is where each test lives. The 2022 European LeukemiaNet recommendations for acute myeloid leukemia, diagnosed from fresh marrow, build a revised genetic risk classification with cytogenetics among its inputs (Döhner and colleagues 2022), an input list in the guideline's tables, not the record I cite. The European Society for Medical Oncology's sequencing recommendations for metastatic solid cancers recommend multigene panels by tumor type and tumor mutational burden testing; their abstract does not mention karyotyping (Mosele and colleagues 2020). The four-society American colorectal guideline recommends mutational testing in the epidermal growth factor receptor pathway (Sepulveda and colleagues 2017); two of that panel carry commercial molecular-diagnostics affiliations.

And the karyotype is being replaced, by something that sharpens the point. Optical genome mapping stretches long DNA molecules out and reads where their landmarks sit, genome-wide, and in 2025 an international consortium recommended it as a standard-of-care cytogenetic assay (Kanagal-Shamanna and colleagues 2025). Verhasselt and colleagues (2025) ran it on 110 non-Hodgkin lymphomas, informative in 94 percent, agreeing with banding and with in situ hybridization, which lights up one named stretch of DNA with a fluorescent probe, and resolving 26 specimens where banding had failed to grow at all. But the landmarks are mapped against a human reference genome, and it still needs cells rather than wax, so the new test asks the old one's question, and a parasite's twelve chromosomes are not in its answer set either.

Now the arithmetic. The dwarf tapeworm has twelve chromosomes in a body cell, from Mutafova and Gergova (1994), published under the name Vampirolepis nana, which is why the modern name alone misses the record. O. viverrini, twelve. Clonorchis sinensis, the Chinese liver fluke, fourteen, on one group's work only (Zadesenets and colleagues 2011). Schistosomes, eight pairs, sixteen, their sex chromosomes running the other way from ours (Stitz and colleagues 2021). Against forty-six. A metaphase spread with twelve chromosomes in it would be the most obvious thing on the slide.

And it never gets the chance. The test that can be run on a wax block reads a list of named human genes. Ross and colleagues (2015) profiled 200 formalin-fixed specimens of carcinoma of unknown primary: at least one genomic alteration in 192 of them, a mean of 4.2 per tumor, TP53, a tumor suppressor, in 55 percent; KRAS, a growth switch stuck on, in 20; MYC, a master switch for growth, in 12. Every author is listed with Foundation Medicine, which sells the assay; the first also lists a medical college. A success for choosing a drug, and silent about non-human material, because no probe is aimed at any.

Which raises the good objection: parasites have those genes too. True, which is not the same as a test being able to see them. A homolog is a related gene; an ortholog the same gene in another species, its letters free to have drifted far; a paralog a duplicate copy inside one species with its own job. Wendt and colleagues (2022) characterized two p53 homologs in a parasitic flatworm by RNA interference: one orthologous to planarian p53 and human TP53, the most-mutated gene in human cancer; the other a parasite-specific paralog required for the response to genotoxic stress, meaning DNA damage. And since p53 is on the table: that one karyotyped fluke-associated cancer did not express p53 at all, in the cell line and the primary tumor alike (Sripa and colleagues 2005). An absent stain is not a mutation — Ross's panel found mutated TP53 in 55 percent, a different measurement — and either way it says nothing about the fluke.

Wendt and Collins (2025) went further, characterizing a schistosome cyclin-dependent kinase inhibitor, a brake on cell division: knocking it and the p53 ortholog down together yielded tumor-like growths, both tumor suppressors in the worm. The brake gene, ubiquitous in parasitic flatworms and absent from their free-living ancestors, they read as possible horizontal gene transfer, a gene moving sideways between species instead of down from a parent. Belyi and colleagues (2009) trace the ancestor of human p53, p63 and p73 to modern sea anemones, over a billion years of conservation.

Two more, specific and old. Osman and colleagues (1999): the Schistosoma mansoni Ras protein, 81 percent identical and 92 percent similar to K-Ras, single-copy. Shoemaker and colleagues (1992): a schistosome homolog of the epidermal growth factor receptor, EGFR, named SER, recovered by homology to the tyrosine kinase domain of erbB — a method that carries its own rebuttal, since they fished with the most conserved stretch of the protein. And this research found no paper verifying a MYC or BRAF ortholog, BRAF being another growth switch in the same chain, in a parasitic worm, so the show claims none.

The counterexample keeps this honest. Ward and colleagues (2004) cataloged 65 protein kinases in Plasmodium falciparum, the malaria parasite, and not one clusters with the tyrosine kinase group, the switches that include EGFR and most famous druggable targets. There is no such target in it to amplify. Worms are animals carrying animal genes; the malaria parasite is far more distant, and for it, at least, the homolog argument does not transfer.

Then the measurement that governs everything. A primer is a short synthetic piece of DNA, about twenty letters, that must stick to the target before anything is copied, and the copying enzyme builds from its three-prime end. Ayyadevara and colleagues (2000) measured what one wrong letter there costs: 40- to 100-fold discrimination against the mismatched template at a three-prime terminal T, G or C, and 8- to 20-fold one base further in. In vitro enzymology with Taq polymerase; and, my inference rather than their measurement, the penalties should compound across both primers. Clinical hotspot assays sit on a handful of exact positions, KRAS codons 12 and 13 among them.

The inference is strong, and here is where it stops. That 81 percent is amino-acid identity, which tolerates enormous difference in the DNA beneath it, because several DNA triplets code for the same amino acid. I could not find a published nucleotide identity for that gene, the number that governs whether a primer anneals. The two must never be converted into each other, and I will not do it.

So, the sentence this beat exists for. No published study I could find has run a clinical cancer panel against purified parasite DNA. That the primers would not bind is an inference from assay design, not a measurement, and I will keep saying it in those words. The experiment is obvious: spike graded quantities of S. mansoni, O. viverrini and Hymenolepis nana DNA into human tumor DNA, run a 500-gene panel and a KRAS, BRAF and EGFR hotspot reaction, report cross-reactivity. Cheap, publishable, not done.

One step is left, and it is the quietest. Sequencing produces reads, which a pipeline lines up against a human reference genome; in pathogen work throwing away the human is step one by design (Yamashita and colleagues 2016), while a tumor pipeline keeps what matches. So what happens to a read that will not line up? Frolova and colleagues (2026) re-analyzed whole-genome sequencing from ten mucinous rectal adenocarcinoma cases to recover microbial reads, and it took an elaborate host-depletion stack — among them the standard human reference, GRCh38 — to clear more than 99.9 percent of human reads. That stack removed only a small fraction of the reads, millions per sample, while simulations kept more than 99.7 percent of the true signal of the Fusobacterium they tracked. Nothing in clinical sequencing requires it, and I could find no audit of how many laboratories run anything like it.

And the cautionary tale, which I tell as my own discipline's. Poore and colleagues (2020) mined 18,116 treatment-naive samples across 33 cancer types from The Cancer Genome Atlas for microbial reads, reported signatures specific to tumor type, and discarded up to 92.3 percent of the sequence data in decontamination. Gihawi and colleagues (2023) re-analyzed it and found, in their abstract's own words, that most of the microbes originally reported as associated with cancer were not present at all in the samples — bacterial calls that were human sequence, and a data-transformation error that manufactured a signature even for microbes with zero reads. Nature retracted the paper in July 2024 (Poore and colleagues 2024); a co-author was affiliated with a commercial genomics company and the paper proposed a microbiome-based diagnostic tool. That is the failure mode a parasite search is most exposed to, and it ran opposite to the hypothesis: human sequence mistaken for foreign.

The contamination that is real is duller. Costello and colleagues (2018) put index swapping, where a read from one patient gets labeled as another's because many samples are barcoded and sequenced together, at 0.2 to 6 percent of reads on 2018 patterned-flow-cell instruments; they offered a remedy, non-redundant dual indexing, and how widely clinical laboratories now use it the record does not say. Zavala and colleagues (2022) add jumping PCR in multiplexed capture, at ancient-DNA input levels far below a clinical specimen. And Salter and colleagues (2014) found contaminating DNA ubiquitous in extraction kits and reagents, varying between batches, and critical in samples of low microbial mass — which is what a parasite search would create.

Which leaves the gap. I could find no published audit of what accredited solid-tumor sequencing laboratories do with their unmapped reads: whether they stay in the alignment file, go to a separate file, or are discarded, and whether any human being ever looks. Absence from a search is not proof of absence, and this episode will keep saying so out loud. The Atlas page on karyotyping and gene panels carries the full account, with the reference-build names.

Two vials and an iodine mount

From the reads nobody reads to the oldest test in the building. The ova-and-parasites exam, ova being eggs, is preserved stool, concentrated, then read twice under a microscope: wet, as a drop under a coverslip, and again as a permanent stained smear, a thin film of stool fixed to a slide and dyed. One Gemini answer says the modern two-vial technique, formalin in one vial and polyvinyl alcohol in the other, was created in 1949 by Brooke and Goldman at the Communicable Disease Center. Each piece has its own date.

The concentration came first. Ritchie (1948) published an ether sedimentation technique for routine stool examinations in the Bulletin of the U.S. Army Medical Department; the record has no abstract, so I name it, date it and quote nothing. Sedimentation spins the stool down so that eggs, larvae and cysts, the walled resting forms of single-celled parasites, settle into a pellet while a solvent carries off fat. The fixative came a year later: Brooke and Goldman (1949), from the Communicable Disease Center's laboratory, titled their note for polyvinyl alcohol as “a preservative and adhesive for protozoa in dysenteric stools”. PubMed does not index it and I have not opened it. Polyvinyl alcohol, PVA, is the synthetic polymer that glues stool to the slide.

Then a single vial. Yang and Scholten (1977) described SAF — sodium acetate, acetic acid and formalin — a simple, stable and, in their words, relatively nontoxic fixative that permitted both concentration and permanent staining. Three ingredients, all in the name; a Gemini document adds phenol, and it is not there. The solvent changed last, ether being volatile: Erdman (1981) concentrated 62 fresh stools in parallel with ether and ethyl acetate and found comparable recovery of eggs, cysts and larvae with no distortion of shape. So a two-vial technique credited to one 1949 team folds thirty-three years, from Ritchie's ether to Erdman's ethyl acetate, into one sentence.

A second Gemini answer makes the preservatives absolute: formalin “completely destroys” trophozoites, the active feeding form, and keeps out trichrome, while PVA distorts helminth eggs into invisibility. The division of labor is real; the absolutes are not. Weber and colleagues (1992), at the Centers for Disease Control, built a stain on the dye chromotrope for microsporidia, one of which, Enterocytozoon bieneusi, causes chronic diarrhea in people with HIV. On unconcentrated, formalin-fixed stool it found the spores in all four patients with biopsy-confirmed infection, and, prospectively, infection in 6 of 27 HIV-infected patients with chronic diarrhea. Formalin-fixed stool took the stain and showed the spores: a substrate, not a ruin.

The PVA preservative carried Schaudinn's fixative, built on mercuric chloride, and mercury disposal had laboratories looking for substitutes. Garcia and colleagues (1983) put a copper sulfate base against the mercuric chloride base in 417 paired specimens: no significant difference in the numbers or shape of organisms in the concentration sediment, where helminth eggs are assessed, but on the trichrome-stained smear copper was not equal to mercury, and they advised against switching unless there was no other option. No comparison with formalin, but helminth eggs in PVA concentrate, there to be counted.

Whether a mercury-free substitute costs anything depends on which one. Jensen and colleagues (2000), at a Veterans Affairs medical center in Atlanta, split 68 fresh stools four ways: mercuric PVA, which they call the “gold standard”, and three mercury-free products, Proto-Fix, Ecofix and Parasafe. Parasites were found in 31, 31, 30 and 30, in that order. Entamoeba histolytica, the disease-causing intestinal amoeba no microscope can tell from its harmless double E. dispar, was seen in 13 Proto-Fix specimens, 7 mercuric PVA, 6 Ecofix and none with Parasafe, which also left a dirty background in 41 percent. Every fixative is a company's product; the record lists only the first author's affiliation.

Flotation is the reverse: in a salt solution denser than they are, eggs rise to the surface film, decided by specific gravity. A Gemini document calls zinc sulfate flotation a “massive blind spot” for heavy eggs, the flukes' among them — laboratory teaching, not a measurement. The one density paper found is David and Lindquist (1982): ten kinds of helminth egg, almost all from animals, on sucrose gradients, from 1.0559 to 1.2376, with no measured density for the human eggs of the dwarf tapeworm, Clonorchis, Fasciola, another liver fluke, or Paragonimus. And the phrase overstates even where the teaching holds: the Centers for Disease Control and Prevention's DPDx guidance, unindexed and reported rather than opened, says operculated and unfertilized eggs do not float well and that a flotation laboratory should read surface film and sediment both; American laboratories concentrate by sedimentation anyway. A blind spot needs a laboratory that floats and looks only at the top. For the dwarf tapeworm it runs the other way: Truant and colleagues (1981), to whom this part returns, found flotation beat sedimentation for its eggs. The same document's claim that fragile Giardia cysts get trapped in the ethyl acetate layer is unverifiable: no study found shows it.

Now the iodine mount, where iodine goes on parasites in order to see them: a drop of stool mixed with Lugol's iodine, iodine dissolved in potassium iodide, under a coverslip, a stain for finding cysts. One thing it costs, from that DPDx guidance: the drop kills trophozoites on contact, so the motility used to recognize them is gone — but that is a live wet mount, and a tumor block was never going to be read for motion. Mewara and colleagues (2019) examined 150 stools unconcentrated, after formol-ether concentration, and after a commercial concentrator, Mini Parasep, reading each by wet mount, iodine mount and a modified acid-fast smear. Positives were 72, 77 and 80, printed as 48.6, 51.3 and 53.3 percent — and the first does not divide: 72 of 150 is 48.0, so the counts go on air and that percentage does not. Sensitivity was 90.1 percent for direct microscopy and 98.7 for Mini Parasep, whose dwarf-tapeworm yield was also the better one. An earlier version of this episode credited the roughly 90 percent, and that yield, to the iodine mount. Backwards: ninety is the worst preparation, the yield is the concentrator's, and the abstract gives no iodine-mount figure at all.

What the iodine mount misses, it misses at low burden. Helmy and colleagues (2009) examined 97 Egyptian stools. Saline wet mount stained with Lugol's iodine called 30, 31 percent, positive for Giardia; PCR aimed at Giardia's tpi gene found 41, 42.3 percent, 11 of them negative under the microscope. Those 11 had higher average cycle thresholds, which the authors read as true, low-burden infections. One limit: the DNA came from the stool, not the stained slide, so this is the microscope running out of parasites, not DNA surviving iodine.

The iodine mount's documented error runs the other way. Uslu and colleagues (2016), in Erzurum, Turkey, took 90 patients called E. histolytica or E. dispar on saline and Lugol's wet mounts. Trichrome failed to confirm the finding in 31.1 percent, and two antigen tests specific to E. histolytica in 62.2 and 64.4 percent, part of that expected, since the two species look identical. In the authors' words, wet-smear microscopy alone “can cause significant false positive results”. Its measured fault is seeing too much. One chemistry footnote: against a Gemini answer that has lone triiodide ions slipping into glycogen's helix, Madhu and colleagues (2016), materials chemists, matched the Raman spectrum of a crystal of endless iodine chains to starch-iodine's, pointing to long polyiodide chains there too. The chain chemistry and glycogen's brown both sit on the Atlas page for the specimen and iodine.

Then the panels, each running many PCR reactions at once, each aimed at a named organism. Two before-and-after studies make the same point. Axelrad and colleagues (2019), at one New York center, found 9,402 patients on the FilmArray gastrointestinal panel positive in 29.2 percent against 4.1 for 5,986 tested conventionally before, with fewer endoscopies, less imaging and fewer antibiotics after; and Beal and colleagues (2017), at the University of Florida, found other infectious stool tests per patient fell from 3.02 to 0.58 across 241 panel patients against 594 prior controls, P = 0.0001, the ova-and-parasites exam not separated out. Associations, one center each. Then the measurement that puts a number on what a panel cannot be asked. Robert-Gangneux and colleagues (2025), at Rennes, ran three years of routine stools through a multiplex panel and microscopy side by side: 3,495 specimens from 2,127 patients. The panel found more protozoa, by a wide margin. Microscopy found helminths, which the panel has no target for, in 68 specimens (63 worms in 60 by its text), and the authors conclude that microscopy is still needed when a worm is suspected. Prospective, one laboratory, routine practice.

Worms can be on a panel. Autier and colleagues (2021), at Rennes University Hospital, evaluated the Allplex GI-Helminth(I) assay, in their title “the first marketed multiplex PCR for helminth diagnosis”, on 110 frozen stools already known positive; its targets include Hymenolepis species and Strongyloides, the threadworm. Bead-beating, breaking the egg open before extraction, made the difference: the whipworm Trichuris went from 1 of 11 to 10 of 11, and Hymenolepis, the dwarf tapeworm's genus, from 20 of 21 to all 21. But it still missed 7 of 13 hookworm, 4 of 11 pinworm, Enterobius, and 7 of 35 Strongyloides, which bead-beating barely helped. The egg has to be cracked first: the shell this part returns to. The maker sold the reagents at a reduced price, and the paper says the assay needs improving before routine use.

Two more results on the same assay. Coppens and colleagues (2024), at the travel clinic of Antwerp's Institute of Tropical Medicine, put 97 stools from 95 patients through that helminth assay: 59.1 percent sensitive against 100 for the clinic's conventional workflow, and they do not recommend it — while the maker's protozoa assay beat that workflow on two targets, 100 percent against 47.4 and 95 against 77.5, and matched it on the pathogenic protozoa, 90 against 95. Better at what it was pointed at, blind to the rest. Köller and colleagues (2020) found the dwarf tapeworm in 40 to 42 of 250 stool DNA samples, the samples picked as likely infected, so that is no prevalence, and one author is from TIB MOLBIOL, a maker of PCR reagents. None of these studies is American or shows clearance by the Food and Drug Administration, so this part's later claim about cleared American panels stays that narrow.

Which test, for whom, is a live document's job. Miller and colleagues (2024) wrote the 2024 update of the Infectious Diseases Society of America and American Society for Microbiology guide to using the microbiology laboratory. I have not read its stool tables, so this episode says nothing about what it recommends; two authors list a diagnostics manufacturer, Cepheid, and a private firm. So the one place a laboratory puts iodine on a parasite on purpose, it does so to see it. Now the claim that iodine destroys it.

The iodine, the panel menus, and the experiment nobody has run

So, the chemistry everybody blames, at full strength. Iodine is on skin before the knife and mucosa before the forceps, and it kills parasites; so a parasite is destroyed before anyone looks, and parasite-driven cancers are systematically missed. Coherent, and nobody's argument in the indexed literature: no PubMed publication states it in any phrasing I searched. It has a popular literature and no scientific one, whose ancestor is a named, dated book, Hulda Regehr Clark's The Cure for All Cancers (1993). PubMed does not index it and I have not opened it, so I describe nothing inside it.

Povidone-iodine, sold as Betadine, does go on skin before an incision — and was beaten in a randomized trial: Darouiche and colleagues (2010), 849 adults having clean-contaminated surgery, chlorhexidine-alcohol cutting surgical-site infection 41 percent against povidone-iodine, relative risk 0.59. Iodine also goes onto surfaces about to be biopsied, as Lugol's solution. Kondo and colleagues (2001) and Jin and colleagues (2019), on 120 and then 240 esophagi, sprayed 10 milliliters of it and quenched the burning. And Morita and colleagues (2017) pooled twelve studies: spraying dye on the gut lining during endoscopy, chromoendoscopy, reached 92 percent per-patient sensitivity for squamous cell carcinoma of the esophagus and for high-grade dysplasia. Poured on tissue about to be sampled, iodine reveals the lesion.

Then the collapse, in five corrections, and the first is against myself. B5 is an old mercury-containing fixative, and clearing its mercury deposits really does cost antibody signal: Wan and colleagues (2003) treated B5-fixed slides with Lugol's iodine and 5 percent sodium thiosulfate before antigen retrieval, and of 75 antibodies stained in paired runs with and without clearing, CD5, CD30 and synaptophysin went blank and three more weakened. That finding stands. What goes is the inference: no record tells me how often mercurial fixatives are still in use, and it is a false negative for antigens, not for a worm's shape. The segment rests on two routes by which iodine reaches a specimen, not three.

Second, povidone-iodine is not uniquely parasite-killing, and that literature never settled. Shi and colleagues (2020): one of four agents after which cysts never recovered on agar in three weeks. Sunada and colleagues (2014), cysts from 56 patients with proven Acanthamoeba keratitis, inflammation of the cornea: natamycin first, povidone-iodine second. Lim and colleagues (2000), nineteen Australian corneal isolates: poor activity. Pelletier and colleagues (2011): povidone-iodine 0.4 percent with dexamethasone 0.1 percent killed 99.9 percent of three bacteria and a Candida in fifteen seconds and a Fusarium at sixty, yet did not inhibit Acanthamoeba castellanii cysts at all. Randag and colleagues (2023), sixteen lens solutions including an iodine-based one: cysts resistant to all. The most favorable result, Bahr and colleagues (2025), is not a cyst result: trophozoites stopped moving inside two minutes at 0.05 percent, a lens-disinfectant strength, and never recovered — and cysts do not move, so that is the same pattern rather than an exception; all five authors list Ophtecs Corporation, a contact-lens-care company. All in vitro, all a free-living amoeba: the tougher the envelope, the less iodine does.

Third, the warning everybody quotes. Heaselgrave and colleagues (2019) cautioned that certain topical anesthetics, and eye preparations preserved with benzalkonium chloride, applied before sampling may cause false-negative tests. The agents named are the anesthetics and the benzalkonium, not iodine — though the same paper ranks povidone-iodine's antiamoebic effect in the dish above the diamidines, the finding against us. And the record says only “diagnostic tests”; the one that needs the organism alive is culture, and cancer specimens are not cultured but fixed. Fourth, the one randomized test. Rajaraman and colleagues (2026) added povidone-iodine to standard chlorhexidine in 49 patients with Acanthamoeba keratitis: 2.5 percent, cut to 1 after reports of intolerance, half-hourly for two days, hourly for four more, then eight times a day out to day 28. At four weeks 6 of 18 were still culture-positive on iodine against 5 of 16 on control; modeled across all three visits, the rate ratio at day 28 was 0.90, interval 0.35 to 2.29, p of 0.82 by a permutation test. The authors call it too small for a definitive answer: the interval allows anything from a 65 percent reduction to a 129 percent increase. A ten-second skin swab will not do what that failed to do.

Fifth, the physics. Brownell and Nelson (2006) called Ascaris eggs, on their own method, the most ultraviolet-resistant water-related pathogen identified to date, and the protection is that shell: stripping its outer layers took the same 500-joule-per-square-meter fluence from 0.44 to 1.80 log inactivation, four times the log kill. Maya and colleagues (2019) record high resistance to conventional disinfection; their inactivation step for Ascaris lumbricoides and Toxocara canis eggs was 60 degrees Celsius for an hour. And killing the embryo inside a shell does not dissolve the shell, which is what a pathologist sees. One measurement is genuinely open: whether iodine at surgical concentration oxidizes parasite DNA enough to defeat a broad-range eukaryotic PCR, eukaryotes being every organism whose cells carry a nucleus.

One more claim needs correcting. “American hospitals never check for Hymenolepis nana”, the dwarf tapeworm, is wrong as stated. Its eggs are found by exactly the exam this part has walked through: preserved stool, concentrated, read under a microscope. Shahnazi and colleagues (2019): 10 of 1,521 Iranian stools by direct wet mount and formalin-ethyl acetate concentration, all ten then sequenced. Abd-Elrahman and colleagues (2024), 120 children with diarrhea in Asyut, Egypt, by direct smear, formol-ether sedimentation and Sheather's sugar flotation: 28.3 percent positive, a DNA method confirming the microscopy at 100 percent sensitivity. Truant and colleagues (1981), more than 250 specimens: zinc sulfate flotation beat sedimentation for this organism. And there is a published American diagnosis, Dalai and Beaven (2020), in an immunosuppressed patient with chronic diarrhea; that record has no abstract, so the fact of it is all I have.

The defensible version is narrower and stronger. Neither cleared multiplex gastrointestinal panel whose menu this part verified, the BioFire FilmArray and the Luminex xTAG, contains any helminth target at all. Buss and colleagues (2015) list all 22 targets of the FilmArray: four are parasites and all four protozoa. No worm; four of its nine authors worked for the manufacturer. Khare and colleagues (2014) record the Luminex xTAG panel as cleared by the Food and Drug Administration for 11 targets, two parasitic; Stockmann and colleagues (2015), describing the FilmArray menu, call its parasite coverage protozoal, the honest word, six of their eleven authors manufacturer employees. So a patient worked up with a panel instead of that exam cannot have a worm found, by design. The panels did win on merit: of Khare's 230 prospective samples, routine testing found 8.3 percent positive, the then-prototype FilmArray 33.0 and the cleared Luminex panel 30.3. But the counterweight belongs in the same breath: Chhabra and colleagues (2026), a Centers for Disease Control verification, re-tested 415 Veterans Affairs stools the panel had called positive for a virus and could not confirm 26.5 percent of its norovirus calls, 21.3 percent of its rotavirus and 18.8 percent of its sapovirus.

One more correction, because the chemistry usually given is wrong. The acid-fast stain works on coccidian oocysts, the thick-walled form these parasites shed, not because their walls are packed with mycolic acids, which nobody reported there, but because, as Bushkin and colleagues (2013) propose, the wall carries a waxy coat of complex triglycerides, and the whole Cryptosporidium oocyst wall dissolves in organic solvents. And the stain is not reliable: Arrowood and Sterling (1989), blinded across 311 stools, had the commercial acid-fast method find 40.6 percent of confirmed Cryptosporidium positives at 52.0 percent specificity, against 100 percent for the two OW3 monoclonal indirect immunofluorescence methods, with no false positives from any monoclonal method.

The old exam has its own reversals, starting with a correction by the name of the error: a positivity rate is not a sensitivity. Positivity depends on who was tested — 6.7 percent of 2,015 exams at Duke (Morris and colleagues 1992), 36.8 percent at a high-prevalence Minneapolis center (Cartwright 1999), same test, different patients. Sensitivity is higher: Branda and colleagues (2006) found 91 percent of parasites in the first specimen submitted, and where three or more came the first alone was 72 percent sensitive, single-specimen rejection then leaving their detection rate unchanged. Morris at Duke stopped examining stool sent after hospital day three, and of 265 rejected specimens the 22 later run on request were negative; Siegel and colleagues (1990) had argued for that rule two years earlier, none of 90 positive exams in three years coming from patients sampled after day three. Against that: Cartwright found the first specimen adequate in only 75.9 percent, and Hiatt and colleagues (1995) at Kaiser found three examinations raised yield over one by 22.7 percent for Entamoeba histolytica and 31.1 for Dientamoeba fragilis.

Another reversal is the preservative. The reference fixative for the permanent stained smear was the mercury-based vial described earlier, and mercury waste rules forced a substitute. Garcia and colleagues (1993) measured the cost: zinc-based polyvinyl alcohol agreed with the mercuric chloride version in 87.0 percent of 161 organism identifications, 92.5 percent once rare organisms were set aside, with pathogens missed in 6 of 161, four amebiasis and two giardiasis, every one at rare numbers. A reversal driven by disposal law, at a measured and modest cost.

One place American practice does name a worm: Strongyloides stercoralis, in patients on steroids, chemotherapy or transplant drugs. Siddiqui and Berk (2001) report that a single conventional stool examination fails to find larvae in up to 70 percent of cases. Centers for Disease Control and Prevention refugee screening guidance, unindexed and not opened for this part, is reported to say that stool examination should not be used to rule it out; the American Society of Transplantation guideline from La Hoz and Morris (2019) screens donors instead. Hence serology — though part four's leukemia patient, larvae filling her lavage, had a negative antibody test. No United States guideline found recommends a routine exam for a cancer patient. The menu argument is current, not a 2015 artifact (Chhabra 2026); what stays half-open is that nobody has measured dwarf tapeworm under-detection here. The argument runs from verified menus, never measured misses.

Nobody has measured how often a tumor specimen holds a parasite nobody named — part four says why no study gives that rate — and feasibility is not the obstacle, though it has a limit. Gomez and colleagues (2017), the fungal sequencing assay part four cited, sequenced two conserved ribosomal regions across 233 fresh and paraffin specimens, and its D2 reads also diagnosed five invasive protozoal infections, a by-product of a fungal assay. The pooled version exists for fungi: Filippidis and colleagues (2026) gathered 28 studies for the Fungal PCR Initiative, and on wax blocks, 18 studies and 852 samples, it reached 75.4 percent sensitivity against 86.5 percent on fresh or frozen tissue. That eleven-point gap is what fixation costs a broad-range reaction, measured — and it is fungi, not worms. Feasibility is otherwise well shown: Bennett and colleagues (2014) assembled a 1.26-gigabase tapeworm genome from one human brain biopsy, and Matsunaga and colleagues (2022) read about 7,000 genes from two-year-old mouse liver blocks and resolved tumor from non-tumor expression in six-year-old human lung cancer blocks. But Grimm and colleagues (2021) put a shelf life on it: their Echinococcus PCR was positive in nine of fifteen samples — all six blocks under six years old, only three of the nine older — and they recommend under five, so a trawl through an old archive is biased toward false negatives.

One frame before the study. The International Agency for Research on Cancer's estimate for 2024, published in 2026, puts the three Group 1 parasites at about 9,700 cancers — 5,900 of the bladder and 3,800 of the bile ducts, figures from the paper's results rather than its abstract, and by my own division about 0.05 percent of all cancer — made by applying attributable fractions to registry counts that record where a tumor is, not what caused it (Rumgay and colleagues 2026); part one derives it and part eight weighs it. The gap has a human number too: among migrant women screened near Barcelona who tested positive, the median was 41 visits to a sexual and reproductive health unit before schistosomiasis was diagnosed at all, on antibody serology (Roure and colleagues 2022).

So here is the study: blinded expert review plus broad-range eukaryotic PCR, across a consecutive series of archived tumor blocks under six years old, negative controls alongside, host sequence removed against more than one reference genome, reporting the fraction that return non-human sequence — likely inheriting the 75.4 percent the pooled fungal version reached on wax — which part eight costs as a protocol somebody could fund. Something like it has been run once, in the wrong organ and with targeted rather than broad-range reactions, the parasite reaction run only on the one case already suspected. And it worked. Hata and colleagues (2026) went through 2,025 consecutive autopsy cases, found five with necrotizing granulomatous lung lesions, walled-off nodules of inflammation dead at the center, trimmed the lesions out of the archival wax by hand and amplified DNA in all five: tuberculosis complex in two, nontuberculous mycobacteria in two, and Dirofilaria immitis in one, by a reaction aimed at cox1, a mitochondrial gene whose sequence names a worm. The Ziehl-Neelsen stain showed abundant bacilli in one tuberculosis case, rare in the other, and nothing in the remaining three, which molecular testing named anyway. A consecutive denominator, old blocks, hand-trimmed lesions, one parasite in five. So the claim, both halves. Iodine did not erase the evidence: added to standard treatment on infected corneas, it left 6 of 18 eyes culture-positive against 5 of 16 without it. And the panels would not have found a parasite anyway, because every one of them is a question about a named human molecule. Two sentences to leave with. A negative report is a statement about the questions asked of the tissue that reached the glass. And the reason we do not know the answer for tumors is that the experiment is cheap, the tissue is already in the building, and nobody has run it.

Sources

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