Collagen and invasion

In brief

What collagen is, and why everything that moves has to deal with it

Collagen is a rope made of three ropes. Each strand coils one way, left-handed, in an open, extended spiral called a polyproline-II helix, and three of those strands wind around each other the other way, right-handed, into a single stiff bundle; the result is the most abundant protein in animals ². The reason the three strands can press together at all is that every third position along each strand is occupied by glycine, the smallest amino acid, which takes up almost no room ^3,36. Written out, the repeat is Gly-X-Y: glycine first, then two positions that can be almost anything, but which in real collagen are most often proline and hydroxyproline — the papers that had to build peptides mimicking collagen write the repeat as Gly-Pro-Xaa ^4,23. Chemists can build the same helix out of nothing but a repeating proline-proline-glycine triplet, and when they replace those prolines with 4-fluoroproline or 4-methylproline the helix becomes the most stable one known: it comes apart only at a temperature more than 50 degrees Celsius above the unmodified version's, and the calorimetry says the gain comes from those substituted side chains restricting how the backbone can twist, so that the chain is pre-set into the shape the helix needs before it folds at all ³⁶. Natural collagen reaches the same end by a different route, by hydroxylating prolines after the chain is made, which is where vitamin C comes in.

There is not one collagen but twenty-eight, counting every member of the family that contains at least one triple-helical domain; four of them sit in the cell membrane and can be shed into a soluble form ¹. Most of the rest are secreted and then assemble further: the fibrillar collagens, type I above all, bundle into the thick fibrils of skin, bone, tendon and tumor stroma, while type IV builds flat sheets instead ^5,24. The number has grown with the sequencing: a standard review in 2003 counted more than twenty ²⁴, which is a small reminder to give the date along with the number. The count has held, but the page should say whose count it is: the figure of 28 comes from that 2011 review and nothing opened for this page revises it, and the current 2026 account of collagen in breast cancer is built on the same family, the same two enzyme classes, the same stiffening and the same signatures without offering a new number ^1,37.

A fibril is not just a heap of collagen molecules. Each one leaves the cell as procollagen, carrying bulky end regions that stop it assembling too early; enzymes outside the cell trim those off, and only then do the rods pack side by side into a cable. The steps of that assembly — the biosynthesis and the supramolecular processing of the fibrillar collagens, type I above all — are the subject of the same 2003 review that supplies the older count ²⁴. What matters for the rest of this page is that the packing is ordered and repeating rather than random, because that order is exactly what makes aligned collagen light up under a laser microscope while the tissue around it stays dark ^28,38.

Two chemical steps turn the chains into usable material. Inside the cell, prolyl 4-hydroxylases add a hydroxyl group to specific prolines, and those hydroxyprolines are what make the finished triple helix stable at body temperature; the enzymes need ferrous iron, 2-oxoglutarate, oxygen and ascorbate to work, which is the whole biochemical content of scurvy ⁴. Outside the cell, the lysyl oxidases, five copper-dependent enzymes, weld neighboring collagen and elastin molecules to one another, and those welds are what make a tissue stiff and resistant to being taken apart again ³².

From the point of view of anything trying to move, all of this is architecture in the way. A cell that stays inside its own compartment never has to touch it. A cell that leaves does, and so does a larva entering through skin, and the physics is the same for both: either widen the gaps, or fit through the gaps that exist ⁷. Which of those two a parasite is doing has, remarkably, never been measured. The parasite literature reports which host proteins get cut ²⁶; it does not report whether the organism had to cut them in order to fit.

The whole history, with every reversal marked

The history of this subject is a history of a good idea getting narrower. The first idea was simple and wrong in its generality: tumors and invading parasites dissolve tissue with enzymes, so block the enzymes. That produced the matrix metalloproteinase inhibitors of the 1990s, which failed in three phase III programs, in the largest pancreatic cancer trial of its day, and left behind a review whose job was to explain what had been misunderstood ^10,29.

What replaced it was narrower and better. The enzyme that mattered turned out to be one membrane-anchored collagenase working in a thin shell against the cell, not a cloud of secreted proteases dissolving tissue at large ⁶. Movement itself turned out not to require enzymes at all in many cells ⁴³, and even where a barrier is breached, part of the work is mechanical shoving rather than chemistry ⁴⁰. Meanwhile the imaging people found that the arrangement of the fibers carried prognostic information, and over fifteen years that grew from a mouse observation to models fitted on nearly a thousand patients ^8,28.

The parasite literature ran on its own track and reversed in its own way: the schistosome enzyme that everybody learned about turned out not to be used by every schistosome, with a different protease family doing the same job in Schistosoma japonicum ⁴⁸. And the most instructive reversal of all came from the pancreas, where removing the collagen, which should by the standard story have helped, made the cancer faster and the survival shorter ⁹. And the parasite literature has since been corrected in the other direction too, by people looking harder rather than by anyone being wrong: the flat claim that no schistosome enzyme had ever been shown to cut collagen turns out to be false for the basement-membrane type, which a purified cercarial proteinase was reported to cut as long ago as 1985 ^49,50, and the fluke cathepsins that were said to lack a collagen assay have had one since 2009 ^23,51.

How a cell gets through tissue: four routes, not one

Getting through packed collagen is a problem with more than one solution, and a cell can use several at once. It can cut the fibers, but only in the narrow space held against its own membrane ⁶. It can avoid cutting altogether and deform itself through gaps that already exist, up to a hard limit set by its nucleus ^7,43. It can push, which is what an invadopodium does when it displaces a basement membrane instead of dissolving it ⁴⁰. It can be towed by somebody else: a cancer-associated fibroblast grips a cancer cell through a direct N-cadherin to E-cadherin junction that strengthens under force and pulls it along ⁶⁷. And it can eat the obstruction ⁴⁶.

The four routes are not equally well established, and they are not equally relevant to patients. The cutting route has the cleanest mechanism and a complete clinical failure behind it ^6,10. The squeezing route has the most precise numbers ⁷. The towing route has been shown with optical tweezers in culture and with the junctions observed in patient material ⁶⁷. The eating route is proven for proline in cells and partly in mice, and not yet proven for collagen itself in a living animal ^46,47.

The sheet that separates staying put from invading

Underneath every epithelium lies a thin, dense sheet of matrix called the basement membrane, and it is built differently from the fibrous collagen of the tissue below. Its architecture is two interlocking networks: laminins, cross-shaped molecules that polymerize into a cell-associated mesh, and type IV collagen, which forms a second network, with the two tied together through nidogen and through the heparan sulfate chains of perlecan and agrin ⁵. That is why type IV collagen is a sheet-former rather than a cable-former, and why the enzymes aimed at it are a different pair from the ones aimed at fibrils ^24,68.

Crossing this sheet is the beginning of the metastatic cascade: local tissue invasion comes first, then entry into the circulation, survival there, exit and colonization of a distant organ ²⁵. The two gelatinases, MMP-2 and MMP-9, are specialized for cleaving type IV collagen and are tied to metastatic progression and worse outcomes in breast cancer, which is why the current generation of drug programs aims selectively at them with small molecules, antibodies, nanobodies and engineered inhibitor proteins instead of blocking every matrix metalloproteinase at once ⁶⁸.

The current account is wider than two enzymes. The same machinery — the matrix metalloproteinases, the membrane-tethered MT-MMPs, the ADAMTS proteases and non-proteolytic regulators — thins, perforates and stiffens basement membranes on purpose during normal development and during wound healing, in transient, tightly placed patterns that guide branching and tissue shape ⁷². In carcinomas the difference is placement and persistence rather than possession: invadopodia concentrate the same enzymes to perforate the sheet, and what is left behind is not inert, with collagen IV escape tracks and cut laminin fragments feeding back as signals for proliferation, survival and new blood vessels ⁷². That is the clearest statement of why blocking these enzymes everywhere was a bad idea: they are the tissue's own repair tools.

But the crossing is not purely chemical. In the clearest live-animal picture available, the invading cell's protrusion displaced basement membrane physically, widening a breach that an enzyme-tipped invadopodium had started ⁴⁰. Parasites make the same point from the other direction. The schistosome larva's secreted protease was seen beside a degraded dermal-epidermal basement membrane during invasion of human skin ²⁷, and a Leishmania surface metalloprotease degrades type IV collagen and fibronectin, with overexpression improving migration through matrix ⁷³. Neither of those experiments shows the enzyme is required; one is morphology, the other is gain of function ^27,73.

Alignment and stiffness: what they predict, and what they do not

The architecture of collagen around a human tumor carries real prognostic information. In 196 breast cancer patients, bundles of straightened fibers running perpendicular to the tumor boundary were associated with poor disease-specific and disease-free survival, with hazard ratios between 3.0 and 3.9 on univariate analysis; separately, the paper reports that the signature held as an independent prognostic indicator regardless of grade, size, estrogen and progesterone receptor status, HER2 status — HER2 being a growth-signal receptor on the cancer cell whose presence changes which drugs a breast cancer is given — node status and subtype. Those are two claims rather than one, and the published abstract gives no confidence intervals and no p-values for either, which is an unusual gap for a prognostic study ³⁵. A later study added five more large-scale signatures and fitted them in 995 patients across a training cohort, an internal validation cohort and an external cohort from a different center, reporting disease-free-survival areas under the curve of 0.838, 0.827 and 0.807 and hazard ratios of 7.03, 6.85 and 4.42, all with intervals ⁸. The most recent work moves the same measurement onto ordinary stained slides by scoring nuclear features relative to the collagen signatures in 941 patients ⁷⁴. The clearest statement that this is about arrangement and not amount comes from outside the breast. In 120 lung adenocarcinomas, 54 of which had spread to lymph nodes and 66 of which had not, fiber alignment measured 0.87 against 0.78 (P=0.003) and elastin volume fraction 0.29 against 0.38 (P=0.001), while the total collagen content of the two groups did not differ at all (P=0.87) ⁶².

Stiffness is the other half of the story, and it is now measured in patients rather than only in gels. Magnetic resonance elastography in 107 colorectal cancer patients found tumor tissue stiffer than normal tissue, correlated with collagen content and with stage ³³; atomic force microscopy found cutaneous, oral and lung squamous carcinomas stiffer than adjacent normal tissue ³⁴. In cells, stiffness is converted into behavior through integrins and focal adhesion kinase, with 60-kilopascal collagen-coated gels increasing proliferation, migration and invasion of anaplastic thyroid cancer cells and focal adhesion kinase inhibition reversing it ⁴², and through the mechanical channel Piezo1, whose activation drives fibroblasts to deposit more collagen and so stiffen the matrix further ³⁴.

None of this has yet been turned into a decision for a patient. Every collagen architecture study is retrospective, and the claim people quote from the 995-patient paper is a model's inference rather than a trial result: its nomogram identified 72 of 357 patients, about one in five, who had unsuccessful five-year disease-free survival and might have been undertreated after surgery ⁸. Every randomized attempt to soften a human tumor's matrix on purpose has failed: simtuzumab against a cross-linking enzyme in 240 patients with pancreatic cancer and 249 with colorectal cancer ^11,12, and losartan against the signals that build stroma in 88 with pancreatic cancer ⁵⁹. The only study that is not a failure is not a success either — a single-arm trial of a lysyl oxidase inhibitor in 24 patients with myelofibrosis, which reports less collagen in the marrow and no outcome at all ⁶⁰. And the direction of effect looks different in different organs, though not as neatly symmetrical as it first appears. The colorectal counterweight is about tumor cells sitting directly against fat cells at the invasion front with no stroma or inflammatory infiltrate between them, and that pattern carried a pooled cancer-specific survival hazard ratio of 1.77 (95% CI 1.55-2.01) across 7 cohorts and 5,932 patients ⁶⁴. That is a finding about missing stroma at a fat boundary, not a measurement of collagen fiber architecture — and in the same organ a high tumor-stroma ratio was adverse across 7,934 patients ⁶⁵ and immature stroma was adverse across 9,849 ⁶⁶. So the colon does not cleanly contradict the breast. What it rules out is any rule as simple as more collagen is worse. A 2025 review states the duality plainly, that collagen both enables invasion and restrains it ²⁵.

How it is found and measured

Collagen is one of the few tissue components that can be imaged without being stained. A pulsed laser passing through aligned collagen generates light at exactly half the wavelength, a process called second-harmonic generation, and because only structures whose molecules are all lined up the same way, with no mirror symmetry to cancel the effect out, can do this, the fibers light up on their own while the tissue around them stays dark ^28,38. That is how the collagen signatures were discovered, in intact unfixed mouse tissue rather than in sections ²⁸. The current generation of the method adds polarization resolution, which reports on molecular orientation inside the fibers, and a set of quantitative descriptors: fiber density, orientation, anisotropy, alignment ³⁸.

The weaknesses are practical and worth naming. Scoring has been observer-dependent, the descriptors are not standardized between laboratories, and the field is now turning to machine learning both to remove manual interpretation and to improve reproducibility, which is itself an admission that reproducibility was a problem ^8,38. The 2026 version of that is a model trained on label-free multiphoton images and on language together, which reads the collagen signatures at a surgical margin from a single unstained section and returns a virtual stained image, a margin heatmap scored pixel by pixel and a written description; it was validated on 158 invasive breast cancer specimens against other pathology vision-language models ⁶³. Note carefully what that is and is not: a reader-assistance tool measured against other software, not a test measured against patient outcomes, and it comes from a group that shares authors with the signature studies it is reading ^8,63,74. Nothing here is a clinical test: no guideline requires a collagen score, and no patient's treatment is chosen by one ⁸.

Two cheaper measurements are closing in, and a third has been sitting there the whole time. The oldest route is also the cheapest, and it is named in the founding human paper itself: that study states that the alignment signature can be scored on routine histopathological sections either by second-harmonic imaging or using picrosirius, an ordinary collagen dye, with no laser at all ³⁵. That no clinical workflow has adopted even that version in fifteen years is part of the story, and it is harder to blame on the cost of instruments. The first of the two newer ones is the ordinary stained slide that every cancer patient already has, read for what the stroma is doing: nuclear features defined against collagen signatures in 941 patients ⁷⁴, or the presence or absence of a stromal reaction at the invasion front in nearly 6,000 ⁶⁴. The second is stiffness, by elastography inside the patient or atomic force microscopy on the excised sample ^33,34. A 2026 single-cell study goes further and proposes that cells carry a transcriptional memory of past stiffness, with stiffened three-dimensional collagen raising expression of RELA and MMP14; that is a hypothesis built from signatures and cell culture, not a measurement in patients ⁷⁵.

• Multiphoton and second-harmonic generation imaging of collagenA laser microscope lights up collagen fibers without any stain, which is how the tumor-associated collagen signatures were defined and scored ^28,38.A research instrument. It is not an ordered clinical test anywhere, and no treatment decision is made from itIn 995 breast cancer patients a signature model gave disease-free-survival AUCs of 0.838 (0.800-0.872), 0.827 (0.779-0.868) and 0.807 (0.754-0.853), with external validation ⁸; the earlier 196-patient study reported hazard ratios of 3.0 to 3.9 with no confidence intervals ³⁵.Every study is retrospective, scoring is observer-dependent, and the quantitative descriptors are still being standardized, now increasingly by machine learning, the 2026 version of which reads a margin from one unstained section and writes the report ^8,38,63. No trial has ever assigned treatment by a collagen score, so its clinical value is unmeasured ⁸.
• Ordinary stained sections read for the stroma (hematoxylin and eosin)The cheapest version of the same question: what the connective tissue at the invasion front looks like on a routine slide ^64,74.This is the slide every cancer patient already has, which is why it is the realistic route to a stromal markerNuclear features defined relative to collagen signatures were independently prognostic in 941 patients, though the paper reports no hazard ratios ⁷⁴. In colorectal cancer, tumor cells sitting directly against fat cells at the invasion front, with no stroma or inflammatory infiltrate between them, carried a pooled cancer-specific survival hazard ratio of 1.77 (95% CI 1.55-2.01) across 7 cohorts and 5,932 patients, and a pooled overall survival hazard ratio of 1.53 (1.39-1.68) across 5 cohorts and 4,545 ⁶⁴.Both readings are retrospective, and the colorectum is not the mirror image of the breast that it first looks like. Dense aligned stroma is bad in breast ³⁵; in the colorectum a high tumor-stroma ratio was adverse across 7,934 patients ⁶⁵ and immature stroma was adverse across 9,849 ⁶⁶, while what the pooled adipocyte analysis scores is tumor touching fat with no stroma in between, which is not a measurement of collagen architecture at all ⁶⁴.
• Tissue stiffness: magnetic resonance elastography and atomic force microscopyElastography measures how hard tissue is to deform inside a patient; atomic force microscopy does it on an excised sample, in kilopascals ^33,34.Elastography is routine for liver fibrosis; for tumors it is still a research measurementIn 107 colorectal cancer patients, tumor stiffness exceeded normal tissue and correlated with collagen content and stage ³³; cutaneous, oral and lung squamous carcinomas measured stiffer than their adjacent normal tissue, and a 53-patient cohort reported a disease-free-survival hazard ratio of 181.03, which is a sign of an unstable estimate rather than a huge effect ³⁴.Stiffness is an association with stage, not a demonstration that stiffness caused it ³³, and every randomized attempt to soften a human tumor's matrix on purpose has failed: simtuzumab against a cross-linking enzyme in pancreatic and then colorectal cancer ^11,12, and losartan against the signals that build stroma in pancreatic cancer ⁵⁹. The one trial that is not a failure is not a success either — an uncontrolled study of a small-molecule lysyl oxidase inhibitor in 24 patients with myelofibrosis reports less marrow collagen and no outcome result ⁶⁰.

What has been tried against it, and what happened

The first generation of drugs blocked matrix metalloproteinases broadly. Phase III trials of marimastat, prinomastat and BAY 12-9566, begun in 1997 and 1998 in advanced lung, prostate, pancreatic, brain and gastrointestinal cancers, reported no clinical efficacy, and two sponsors shut their programs down ²⁹. The largest of those trials randomized 414 patients with unresectable pancreatic cancer to three doses of marimastat or to gemcitabine and found no significant survival difference (p = .19), with median survivals of 111, 105, 125 and 167 days and musculoskeletal toxicity in 44% of marimastat patients ¹⁰. The post-mortem identified three errors: most tumor matrix metalloproteinases come from stromal cells, cancer cells induce them in those cells rather than making them, and some of these enzymes generate inhibitors of angiogenesis, so blocking everything blocks the brakes as well ²⁹.

The second generation went after the cross-linking enzymes instead. Simtuzumab, a humanized antibody against lysyl oxidase-like 2, was added to gemcitabine in 240 patients with metastatic pancreatic adenocarcinoma: median progression-free survival was 3.7, 3.5 and 3.7 months, with hazard ratios of 1.09 (95% CI 0.74-1.61, p = .73) and 1.13 (0.76-1.66, p = .61), and objective response rates of 13.9% and 14.5% against 23.5% on placebo ¹¹. A trial co-author was an employee of the company developing the antibody ¹¹. The same antibody was tested a second time, in 249 patients with KRAS-mutant metastatic colorectal adenocarcinoma given FOLFIRI chemotherapy plus simtuzumab at one of two doses or placebo. Median progression-free survival was 5.5, 5.4 and 5.8 months, with adjusted hazard ratios of 1.32 (95% CI 0.92-1.89, p = .10) and 1.45 (1.01-2.06, p = .04) against placebo, so the lower dose came out significantly worse, and median overall survival was 11.4 and 10.5 months against 16.3 on placebo ¹². Two randomized trials of one antibody, in two cancers, both pointing the wrong way. Defenders of the target argue the antibody was inefficient rather than the enzyme wrong ³², and a 2026 review of a different family member concedes that clinical development is still limited by heterogeneity, missing biomarkers and compensation within the family ⁶¹.

The third generation is narrower by design, and it has reached patients once. Its rationale is selective inhibitors, monoclonal antibodies, nanobodies and engineered inhibitor proteins aimed at MMP-2 and MMP-9 specifically rather than at the whole family ⁶⁸; antibodies against the external domain of the collagen receptor DDR1, which disrupted collagen fiber alignment, relieved the immune exclusion it causes and inhibited growth in mice ⁴¹; and collagen-binding integrins such as α11β1, validated so far in arthritic mice rather than in cancer ⁷¹. The one selective agent that has completed a randomized trial failed. Andecaliximab, an antibody against MMP-9, was added to chemotherapy in 432 patients with advanced gastric or gastroesophageal junction adenocarcinoma: median overall survival was 12.5 against 11.8 months and median progression-free survival 7.5 against 7.1, and although the response rate was higher on the antibody, 51% against 41%, the trial did not meet its primary endpoint ¹³. One second-generation drug has reached patients too. A small-molecule inhibitor of all the lysyl oxidases was given to 24 patients with relapsed or refractory myelofibrosis, which is itself a neoplasm of blood-forming cells: 200 mg twice daily was tolerated, systemic lysyl oxidase activity fell, 13 of the 24 completed 24 weeks, and the authors report early signs of effect including reduced collagen in the marrow ⁶⁰. That study was single-arm and uncontrolled, so nothing in it is a demonstrated benefit, and nine of its authors work for the developer. The honest summary is therefore narrower than it used to be: no randomized trial has ever improved a cancer outcome by acting on collagen or its enzymes, one uncontrolled study suggests the chemistry can at least be moved inside a patient, and one experiment suggests that succeeding too well could hurt, because removing collagen from the pancreatic stroma accelerated the disease ^9,60.

What was believed, and is not

This field has corrected itself unusually openly, and most of the corrections came from inside it rather than from critics outside. What follows is the list of beliefs that have been given up, each with the result that ended it and the year that result appeared. None of them were foolish beliefs. Several were the obvious reading of the best evidence available at the time, which is the point: a mechanism can be correct in a dish, reproducible in mice, and still wrong about what to do for a patient ^9,29,43.

Two of them are worth separating, because they failed in different ways. The claim that blocking lysyl oxidase eliminates metastasis failed because the paper itself was withdrawn, fourteen years after publication, so there is nothing left to weigh ^14,15. The claim that schistosomes get through skin with cercarial elastase failed in the ordinary way, by being tested in more species: the enzyme is there in Schistosoma mansoni, as an expanded gene family, and the same job is done by a different protease family in S. japonicum ^48,76. One is a withdrawal; the other is science working.

The parasite side: who actually cuts connective tissue

Parasites that enter through skin or gut wall do carry tissue-degrading enzymes, and in a few cases the evidence is good. The strongest is a hookworm metalloprotease: a secreted astacin-like enzyme, Ac-MTP-1, digested gelatin, collagen, laminin and fibronectin, and antiserum against it cut collagen digestion by 85% and larval migration through tissue by 70-75%, against 5-10% for preimmune serum, while metalloprotease inhibitors reduced skin penetration by 43-61% ²⁰. That is an enzyme with a substrate, a localization, a loss-of-function test and a migration readout. It is also the molecule the authors are developing as a vaccine antigen, which is a stake worth stating ²⁰.

The schistosome story is better known and weaker than its reputation. A serine protease is made in the acetabular glands of Schistosoma mansoni cercariae and was seen being released into human skin, next to degraded epidermal cells and degraded dermal-epidermal basement membrane ²⁷. In S. mansoni that enzyme is an expanded gene family, apparently unique to the species and probably a matter of dosage, and it is already switched on before the larva leaves the snail ⁷⁶. But S. japonicum does not appear to use it: its glands showed 40-fold greater cathepsin B-like activity instead ⁴⁸, an elastase isoform was undetectable there by antibody ⁷⁷, and when the two enzyme types were compared on human skin they cut many of the same substrates, so two unrelated families solve one problem ²⁶. One more correction to the reputation, and it cuts the other way. The same enzyme has a second job at the skin interface that is better supported than its collagen-cutting one: it interferes with how host cells present parasite antigens and with antibody function, which makes it an immune-evasion molecule as much as a burrowing tool ⁸⁰. So the deflation is specific, not general. Cercarial elastase is real and consequential; what it is not is a general collagenase, and it is not what every schistosome uses to get in ^26,48,80.

The measurements that do exist are mostly on substrates chosen for convenience rather than on the barrier itself. A refolded recombinant S. japonicum elastase is measurably active on an artificial substrate: the concentration at which it works at half its top speed, the Michaelis constant, came out at 0.116 millimolar — the PubMed record garbles the symbol for it — and its top speed was 1,900 nanomoles of product per minute per milligram of enzyme, the product being a dye released from the synthetic peptide when the enzyme cuts it ⁸¹. Both numbers describe the enzyme meeting a short synthetic peptide in a tube, not skin and not collagen. A bird schistosome's cathepsin B was characterized on synthetic peptide substrates at an optimal pH of 6.0 and separately showed elastin-cutting activity, which its authors propose substitutes for the serine elastase other species use ⁸². None of those modern papers reports activity against purified collagen. But the categorical version of that sentence is wrong, and it is worth saying exactly how. The one purified schistosome enzyme that was put on collagen, forty years ago, behaved as a type-specific collagenase against the basement-membrane collagens IV and VIII while leaving the interstitial collagens I, III and V alone ⁴⁹, and the purified cercarial elastase itself degraded type IV collagen along with elastin, gelatin, laminin, fibronectin and keratin ⁵⁰. So the phrase schistosome collagenase is right for the sheet collagen of a basement membrane, unsupported for the fibrillar collagen of dermis, and in both cases resting on purifications from 1985 that nobody has repeated with modern tools ^{26,49,50,81,82}.

Among the protozoa, one result stands out because it involves the same molecules as the cancer story. In human colon explants, Entamoeba histolytica could not invade when matrix metalloproteinase activity was blocked, and the parasite's own cysteine proteinase A5 activated human pro-MMP-3, which in turn activated pro-MMP-1, the host's own collagenases, to remodel fibrillar collagen ²¹. So the amoeba does not invent a collagenase of its own for this job; it switches on the human enzymes that tumors also switch on. There is no tumor anywhere in that experiment — the shared thing is the host's MMP-1 and MMP-3, which the two invaders turn to the same use ²¹. A 2026 version of the experiment uses thin living slices of human intestine and follows the parasite's own genes as it goes in: the adhesion lectin holds a steady level while the cysteine protease genes switch on early, as the amoeba moves from damaging the epithelium to entering the connective tissue beneath it, and stress-response genes rise later ⁸³. The readouts there are gene expression and histology rather than collagen cleavage, so it supports the timing of the protease story without adding a collagen measurement. The amoeba's own collagenolytic activity is also older and better established than this page used to say. It was quantified in 1982, when trophozoites were shown to digest three and a half times more labeled type I than type III collagen ⁵²; it was correlated with strain virulence by an independent group in 1983 ⁵³; and it was measured at 1.66 units per milligram of protein in the parasite's electron-dense granules in 1990 ⁵⁶. What has never been confirmed is the 1992 assignment of a 30-kilodalton collagen receptor ⁵⁵. For Trypanosoma cruzi, the cause of Chagas disease in 6 to 7 million people in Latin America, the claim that the parasite degrades host collagens reaches most readers through a review ⁸⁴ — but the primary paper does exist after all: a secreted 80-kilodalton proteinase, purified from the forms that infect mammalian cells, hydrolyzed purified human collagen types I and IV at neutral pH while sparing albumin, laminin and immunoglobulin, and cut native type I collagen in rat mesentery ⁷⁸. And for Leishmania the surface metalloprotease gp63 degrades type IV collagen and fibronectin, with overexpression improving migration ⁷³.

Then there are the parasites that do not cut at all. Toxoplasma and Plasmodium move by gliding: myosin anchored to an inner membrane complex drags surface adhesins backward along short, rapidly turning-over actin filaments, which pushes the parasite forward with no requirement to digest the surroundings ^45,85. The coupling is now structural. The glideosome-associated connector, the conserved protein that links those actin filaments to the surface adhesins and carries the motor's force to the ground, was crystallized in 2023: a supercoiled armadillo-repeat arm that closes into a ring, with solution data suggesting it also opens into extended forms, which is how one protein could both hold and release ⁸⁶. And the motor is gated rather than standing permanently ready — the rings at the parasite's tip are the hub where glideosome components gather, and extruding the cone-shaped apical structure directs the flux of actin filaments to the surface and so controls motility and invasion ⁸⁷. That is a complete force-transmission chain with no protease anywhere in it. Malaria sporozoites must migrate through the dermis to find a blood vessel, and that movement is the first obligatory step of human infection ⁸⁸. Imaging them inside living skin shows two gears: a fast, far-ranging search of the kind that is mathematically efficient for finding scarce targets, and, on touching a vessel, a slow local probing mode concentrated at entry sites marked by pericytes, the cells wrapped around small vessels ⁸⁹. In engineered pillar arrays they also preferentially circled pillars matching their own curvature, navigating by shape rather than by chemistry ⁹⁰. No matrix-cutting enzyme appears anywhere in that account. Toxoplasma makes the point from inside a host cell: infected dendritic cells crossed three-dimensional collagen by integrin-independent amoeboid movement, and blocking five different integrins had no significant effect ⁴⁴.

The fluke cathepsins deserve more than one careful sentence now, because the experiment they needed has been done. Newly excysted juvenile Fasciola hepatica, the stage that bores through the intestinal wall, expresses its own set of cathepsins L and B, and prefers substrates with proline in the second position, a preference otherwise shared with vertebrate cathepsin K ⁹¹. Cathepsin K is the human enzyme that does cut collagen, inside fibroblasts after fibrils have been taken up, and inhibitors of cathepsins L, B and S do not replace it ⁹². The collagen assay that preference pointed to followed. FhCL3, the main enzyme secreted by the invasive juvenile, cleaved native type I collagen efficiently across a range of pH and temperature while leaving immunoglobulin alone ⁵¹, and FhCL2 and FhCL3 cut native type I and II collagen at multiple sites inside the triple-helical regions of the α1 and α2 chains, while FhCL1 cut only the non-collagenous parts ²³. What is missing is no longer the assay but the next experiment: nobody has taken these enzymes away from a fluke and watched whether it can still get through a gut wall. For Strongyloides stercoralis, which penetrates skin directly, reinfects its host and can kill immunosuppressed people ⁹³, the enzyme has a name as well. Infective larvae secrete a histolytic zinc metalloprotease with elastase activity that degrades a model of dermal matrix, and metalloprotease inhibitors prevented larvae from invading skin in vitro ⁹⁴; the activity was sized as a 40-kilodalton zinc endopeptidase and named Ss40 ⁹⁵, and cloned as strongylastacin, an astacin-like metalloprotease transcribed only in the infective larva ⁹⁶. Collagenolytic activity was measured in Strongyloides infective larvae as early as 1987, although in acidic homogenates rather than in secretions at the pH of skin ⁹⁷. The 1990 substrate comparison, in which Strongyloides degraded elastin, laminin and fibronectin where hookworms degraded only fibronectin, is therefore not the only thing on the record, and what is still absent is a knockout rather than a first result ⁵⁴.

What is genuinely shared, and what is only analogy

Three things are genuinely shared at the molecular level. First, the substrate: both a tumor cell and a larva meet the same two materials, a fibrillar collagen scaffold and a type IV collagen sheet, built the same way in every vertebrate host ^5,24. Second, the strategy of pericellular, localized cutting rather than wholesale dissolution: the tumor cell does it with a membrane-tethered collagenase in the shell against its own surface ⁶, and the hookworm larva does it with enzymes released from gland ducts opening at its own surface and cuticle ²⁰. Third, and most concretely, the host's own proteases — in two parasites now. Entamoeba histolytica switches on human pro-MMP-3 and then pro-MMP-1 to remodel colonic collagen, which is the tumor's own machinery turned to the same use by a parasite; and it is worth being exact about what that sentence means, because there is no tumor anywhere in the experiment, and the shared thing is the host enzyme rather than anything passing between the two invaders ²¹. And Fasciola hepatica juveniles, the stage that bores through the intestinal wall, recruit host plasminogen and generate plasmin in the space immediately around themselves, which came with more collagen degradation and more urokinase secretion by the intestinal cells beside them, and blocking fibrinolysis with a drug reduced the juveniles' migration in mice ²²; a surface enolase on the fluke does double duty as the plasminogen receptor and a laminin-binding adhesin ⁵⁸. There is an irony worth stating. Plasmin is precisely the protease the gene-deletion experiments ruled out for tumor cells, which found MT1-MMP instead ⁶. Both organisms keep the cutting to a thin shell around themselves; they simply conscript different host enzymes to do it.

Almost everything else in the comparison is analogy. There is no parasite MT1-MMP. There is no parasite enzyme known to cross-link host collagen and no parasite equivalent of the lysyl oxidases acting on host tissue, no parasite tumor-associated collagen signature and no parasite mechanotransduction story; nothing of the kind turned up in the searches run for this page, and the stiffening half of the picture is cancer-only and has to be presented that way ^31,32,34. But it would be wrong to say that parasites have no collagen chemistry of their own, and leaving it out hides the sharpest resemblance available here: the invader is built from the same molecule as the barrier. Nematodes are wrapped in a cuticle built of collagens. In Caenorhabditis elegans — a free-living nematode, a relative of the parasitic worms rather than one of them — that cuticle collagen matures outside the cell by dityrosine cross-linking rather than by a lysyl oxidase, and the worm's own prolyl 4-hydroxylase depends on ascorbate exactly as the human enzyme does, so a deficiency that stripped the worm of its ascorbate lowered its collagen and raised its cross-linking ⁹⁸. That belongs in the shared column, with a caveat: the thing crossing the barrier is built from the same kind of molecule as the barrier, with the same vitamin dependency in its making, and only the cross-linking chemistry differs — but it has been shown in a worm that infects nobody, and nobody has done the same experiment in a hookworm or a Strongyloides ⁹⁸. Nor is there a single experiment that put a parasite and a tumor cell in the same assay and compared them. The entire synthesis is cross-literature inference between fields using different organisms, different models and different readouts ^6,20.

There is one more asymmetry worth stating, because it runs the other way. Several parasites get through tissue without cutting anything at all: the gliding apicomplexans move by dragging adhesins backward with myosin ^45,85, malaria sporozoites navigate the dermis by curvature ⁹⁰, and Toxoplasma-infected host cells cross collagen amoeboid-style with integrins blocked ⁴⁴. Cancer cells and leukocytes can do the same thing, within a measured physical limit ^7,43. If there is a real convergence between parasites and tumors in tissue invasion, the best candidate is not shared enzymology at all; it is this: both know when not to cut ^7,44 — and on the cancer side that is not a quirk but a recognized cell state with its own machinery, its own triggers and its own clinical signature ⁶⁹.

What is still unknown

Whether any of this changes an outcome. Every human study of collagen architecture is retrospective prognostication, and the suggestion people quote from the 995-patient study is a model's inference: its nomogram identified 72 of 357 patients, about one in five, who had unsuccessful five-year disease-free survival and might have been undertreated ⁸. The settling study is obvious and has not been done: randomize adjuvant treatment by a collagen score against standard clinicopathologic assignment ^8,74.

Whether MT1-MMP-dependent collagenolysis is rate-limiting in a human tumor. All of the evidence is cell-based or mouse, and there has never been a human intervention against that enzyme at all; the one selective program against a gelatinase that did reach a randomized trial, an antibody against MMP-9 in 432 patients, missed its primary endpoint ^6,13,68. Whether tumors really eat collagen in a living animal is also open: proline from collagen feeds pancreatic cancer cells and proline oxidase is required in vivo ⁴⁶, but the in vivo protein-drinking experiment tracked albumin ⁴⁷, so no one has yet followed labeled collagen into tumor biomass.

Which direction stromal collagen pushes, in which organ. Density caused mouse mammary tumors ⁵⁷; alignment predicts worse human breast survival ^8,35 and tracks nodal spread in lung with no difference in collagen quantity at all ⁶²; deleting collagen accelerated mouse pancreatic cancer ⁹; and in the colorectum a high tumor-stroma ratio, immature stroma and absent stroma at a fat boundary have each been reported as the adverse finding, across 7,934, 9,849 and 5,932 patients respectively ^64,65,66. No study reconciles these, and a 2025 review can only name the duality rather than resolve it ²⁵. Part of the answer may be that stroma is not one thing and fibroblasts are not one cell type, which single-cell work is now making explicit, down to a fibroblast state that restrains a tumor rather than helping it ⁷⁰. Whether the lysyl oxidase family is a wrong target or was merely attacked with an inefficient antibody is likewise unresolved, and the people who believe the latter say so in print ^32,61.

On the parasite side the gaps have moved, and several of them have closed. No paper shows a schistosome enzyme cutting purified type I collagen; the one purified cercarial proteinase that was tested behaved as a type-specific collagenase on the basement-membrane collagens IV and VIII and not on the interstitial collagens I, III and V, and that result is from 1985 and has never been repeated with modern tools ^49,50. The fluke cathepsin-collagen link no longer rests on a shared substrate preference: FhCL2 and FhCL3 cleave native type I and II collagen inside the triple helix ^23,51. What is missing there is a loss-of-function test during invasion. Strongyloides has a named invasion metalloprotease and inhibitor evidence that it is needed to enter skin ^94,95,96, but no knockout and no assay on purified collagen from a secreting larva ⁹⁷. The Trypanosoma cruzi collagen claim has a primary paper behind it after all, with a clean specificity in it, but no loss-of-function test either ⁷⁸. The 1992 amoebic collagen receptor has never been replicated, although the collagenolytic activity around it was reported a decade earlier and reproduced independently ^52,53,55. And the retraction audit this page called for has a result. The lysyl oxidase lineage carries not one mark but three: the 2006 Nature paper retracted in 2020 ^14,15, the same group's 2015 Nature paper retracted in 2023 ^16,17, and its 2013 Cancer Res paper under an expression of concern since 2019 ^18,19. Nothing further should be built on that lineage without saying so out loud. Apart from those, no retraction and no expression of concern appeared on any of the records opened for this page.

How it fits the season's question

The season asks whether the cancers attributed to parasites are the real number or only the number anyone has looked for. Collagen is where that question becomes physical rather than statistical. A tissue that has been crossed, scarred, repaired and crossed again by a migrating larva is a tissue whose scaffold has been rebuilt, and the cancer literature has shown, in mice, that a denser and more cross-linked scaffold can change how an epithelium behaves ^31,57. That is a real mechanism and a plausible bridge, and one coarse version of it has actually been measured in people. In regions where the liver fluke Opisthorchis viverrini is endemic, periductal fibrosis — scarring of the tissue around the bile ducts — seen on ultrasound was pooled against cholangiocarcinoma-related outcomes across 3 studies and 758,686 participants, giving an odds ratio of 2.77 (95% CI 2.24-3.44) with no heterogeneity between the studies ⁹⁹. The authors are explicit that outcome definitions and adjustment varied between those studies, that this is a summary association rather than a risk estimate, and that fibrosis on ultrasound is neither a histological diagnosis nor a proven causal precursor ⁹⁹. So the far side of the bridge is not empty: a gross fibrosis phenotype in parasitized tissue does track a cancer endpoint. What has never been done is the fine-grained version. Nobody has measured collagen fiber architecture — the aligned-fiber patterns that carry hazard ratios of 4 to 7 in breast tissue — in tissue chronically infected by a parasite, and followed those people to cancer ^8,99.

The discipline the page is meant to teach is the one the field learned the hard way. The mechanism was beautiful, the imaging was beautiful, the hazard ratios were large, and the drugs did nothing: 414 patients on marimastat, then 240 and 249 on simtuzumab, then 432 on an antibody against MMP-9, then 88 on losartan ^{10,11,12,13,59}. The founding paper of the stiffening story was retracted, and two more papers from the same group have since been withdrawn or flagged ^15,17,19. And when somebody finally removed the collagen from a pancreatic tumor, the cancer got worse ⁹. If a parasite-cancer argument is going to lean on matrix remodeling, it has to carry those facts with it ^9,10,11,15.

What can be said plainly is narrower and still useful. Parasites really do secrete enzymes that take connective tissue apart, and in one case the loss-of-function evidence is good ²⁰. Two parasites really do conscript the host's own proteases — the amoeba the host's collagenases ²¹, the liver fluke juvenile the host's plasmin, with a drug experiment in mice behind it ²² — and that is the genuine molecular overlap with tumor invasion on this page. Several parasite enzymes do cut collagen, with the type of collagen stated and the year stated ^23,49,78. The rest of the resemblance is analogy between two literatures that have never been tested in the same experiment ^6,20. Saying exactly that, and no more, is what the rest of the season has to be built on.

Where it connects

Sources

99 sources, numbered as they are cited. Every one was checked against PubMed or its publisher before it was cited here; the note under each says what it shows and what it does not.

1. 1
   Ricard-Blum S. The collagen family.Cold Spring Harb Perspect Biol · 2011 · 3(1):a004978doi:10.1101/cshperspect.a004978 · PMID 21421911

   The source of the count of 28 collagens containing at least one triple-helical domain, four of them type II membrane proteins that can be shed into soluble form. A review; it gives no prevalence or outcome numbers.

2. 2
   Shoulders MD, Raines RT. Collagen structure and stability.Annu Rev Biochem · 2009 · 78:929-58doi:10.1146/annurev.biochem.77.032207.120833 · PMID 19344236

   The physical chemistry of the molecule: collagen as the most abundant protein in animals, a right-handed bundle of three left-handed polyproline-II-type helices, with stability explained by stereoelectronic effects rather than by hydration alone. A review of structure; it contains no disease or cancer data.

3. 3
   Marini JC, Forlino A, Cabral WA, et al. Consortium for osteogenesis imperfecta mutations in the helical domain of type I collagen: regions rich in lethal mutations align with collagen binding sites for integrins and proteoglycans.Hum Mutat · 2007 · 28(3):209-21doi:10.1002/humu.20429 · PMID 17078022

   The mutation census: 832 structural mutations in COL1A1 and COL1A2, of which 682 substitute for a glycine in the triple-helical domain and 150 alter splice sites, with lethal clusters aligning to ligand-binding regions. A curated database review, so it describes which mutations were submitted, not their frequency in the population.

4. 4
   Myllyharju J. Prolyl 4-hydroxylases, key enzymes in the synthesis of collagens and regulation of the response to hypoxia, and their roles as treatment targets.Ann Med · 2008 · 40(6):402-17doi:10.1080/07853890801986594 · PMID 19160570

   States that the 4-hydroxyproline residues made by the collagen prolyl 4-hydroxylases are essential for triple-helix stability, and that all prolyl 4-hydroxylases require ferrous iron, 2-oxoglutarate, oxygen and ascorbate. A review of enzymology and drug targets; it reports no clinical trial results of its own.

5. 5
   Hohenester E, Yurchenco PD. Laminins in basement membrane assembly.Cell Adh Migr · 2013 · 7(1):56-63doi:10.4161/cam.21831 · PMID 23076216

   The architecture of the barrier: a cell-associated laminin network and a second network of type IV collagen, linked through nidogen and the heparan sulfate chains of perlecan and agrin. A structural review; it says nothing about cancer.

6. 6
   Sabeh F, Ota I, Holmbeck K, et al. Tumor cell traffic through the extracellular matrix is controlled by the membrane-anchored collagenase MT1-MMP.J Cell Biol · 2004 · 167(4):769-81doi:10.1083/jcb.200408028 · PMID 15557125

   The cleanest mechanism in the field: with gene-targeted cells, crossing fibrillar collagen required MT1-MMP and not plasminogen, MMP-2/TIMP-2, MMP-9, MMP-13, MMP-8 or MMP-3. Cell-based and mouse work; no human intervention has ever tested it.

7. 7
   Wolf K, Te Lindert M, Krause M, et al. Physical limits of cell migration: control by ECM space and nuclear deformation and tuning by proteolysis and traction force.J Cell Biol · 2013 · 201(7):1069-84doi:10.1083/jcb.201210152 · PMID 23798731

   Puts numbers on the limit: protease-independent migration fell linearly with pore size and stopped when the nucleus was squeezed to about 10% of its cross section, 7 square micrometers for tumor cells, 4 for T cells and 2 for neutrophils; beyond that, movement required MMP cleavage to widen the pores. Engineered matrices in vitro.

8. 8
   Xi G, Guo W, Kang D, et al. Large-scale tumor-associated collagen signatures identify high-risk breast cancer patients.Theranostics · 2021 · 11(7):3229-3243doi:10.7150/thno.55921 · PMID 33537084

   The best-reported human collagen prognostic dataset: 995 patients in training, internal and external cohorts, disease-free-survival AUCs of 0.838 (0.800-0.872), 0.827 (0.779-0.868) and 0.807 (0.754-0.853) and hazard ratios of 7.03 (4.87-10.16), 6.85 (4.37-10.73) and 4.42 (2.92-6.71). Retrospective prognostication; its claim that the nomogram identified 72 of 357 patients with unsuccessful five-year disease-free survival who might have been undertreated is a modeling inference, not a trial result, and the abstract's wording leaves it ambiguous what that denominator of 357 counts.

9. 9
   Chen Y, Kim J, Yang S, et al. Type I collagen deletion in αSMA+ myofibroblasts augments immune suppression and accelerates progression of pancreatic cancer.Cancer Cell · 2021 · 39(4):548-565.e6doi:10.1016/j.ccell.2021.02.007 · PMID 33667385

   The experiment that reversed the direction of effect: deleting type I collagen in pancreatic myofibroblasts accelerated tumor development and shortened survival through CXCL5-driven myeloid recruitment, which makes stromal collagen restraining as well as enabling. A genetic mouse model; no effect sizes in the abstract.

10. 10
    Bramhall SR, Rosemurgy A, Brown PD, Bowry C, Buckels JA. Marimastat as first-line therapy for patients with unresectable pancreatic cancer: a randomized trial.J Clin Oncol · 2001 · 19(15):3447-55doi:10.1200/JCO.2001.19.15.3447 · PMID 11481349

    The largest randomized trial of a matrix metalloproteinase inhibitor in pancreatic cancer: 414 patients, no significant survival difference across marimastat doses and gemcitabine (p = .19), median survivals of 111, 105, 125 and 167 days, and musculoskeletal toxicity in 44% of marimastat patients. The authors still read a dose response into it and called for more studies.

11. 11
    Benson AB, Wainberg ZA, Hecht JR, et al. A Phase II Randomized, Double-Blind, Placebo-Controlled Study of Simtuzumab or Placebo in Combination with Gemcitabine for the First-Line Treatment of Pancreatic Adenocarcinoma.Oncologist · 2017 · 22(3):241-e15doi:10.1634/theoncologist.2017-0024 · PMID 28246206

    The randomized human test of blocking a lysyl oxidase: 240 patients, progression-free survival hazard ratios 1.09 (95% CI 0.74-1.61, p = .73) and 1.13 (0.76-1.66, p = .61), with response rates numerically worst in the treated arms. A co-author was an employee of Gilead Sciences, which was developing the antibody.

12. 12
    Hecht JR, Benson AB, Vyushkov D, Yang Y, Bendell J, Verma U. A Phase II, Randomized, Double-Blind, Placebo-Controlled Study of Simtuzumab in Combination with FOLFIRI for the Second-Line Treatment of Metastatic KRAS Mutant Colorectal Adenocarcinoma.Oncologist · 2017 · 22(3):243-e23doi:10.1634/theoncologist.2016-0479 · PMID 28246207

    The second randomized cancer trial of the same anti-lysyl-oxidase-like-2 antibody, published alongside the pancreatic one: 249 patients, median progression-free survival 5.5, 5.4 and 5.8 months with adjusted hazard ratios of 1.32 (95% CI 0.92-1.89, p = .10) and 1.45 (1.01-2.06, p = .04) against placebo, and median overall survival 11.4 and 10.5 months against 16.3. Numerically worse than placebo throughout and significantly worse in one arm, which is stronger evidence against the target than a null result would be. A co-author was an employee of Gilead Sciences, which was developing the antibody. The PubMed record drops the italicized gene name from the title.

13. 13
    Shah MA, Bodoky G, Starodub A, et al. Phase III Study to Evaluate Efficacy and Safety of Andecaliximab With mFOLFOX6 as First-Line Treatment in Patients With Advanced Gastric or GEJ Adenocarcinoma (GAMMA-1).J Clin Oncol · 2021 · 39(9):990-1000doi:10.1200/JCO.20.02755 · PMID 33577358

    The selective-enzyme generation's one completed randomized trial, and it failed: 432 patients randomized 1:1 to chemotherapy with or without andecaliximab, an antibody against MMP-9, with median overall survival 12.5 against 11.8 months and median progression-free survival 7.5 against 7.1 months. The response rate was numerically higher on the antibody, 51% against 41%, and the apparent benefit in patients aged 65 or over is explicitly unadjusted for multiplicity. Three authors were employees of Gilead Sciences, the sponsor.

14. 14
    Erler JT, Bennewith KL, Nicolau M, et al. Lysyl oxidase is essential for hypoxia-induced metastasis.Nature · 2006 · 440(7088):1222-6doi:10.1038/nature04695 · PMID 16642001

    RETRACTED. As published it claimed that inhibiting lysyl oxidase eliminates metastasis in mice and that high-LOX tumors predicted poor survival. PubMed carries it as a Retracted Publication, so none of it counts as evidence. Its first author appears again on a 2015 Nature paper retracted in 2023 and on a 2013 Cancer Res paper under an expression of concern, and both she and its senior author are co-authors of the 2009 Cell cross-linking paper that survived.

15. 15
    Erler JT, Bennewith KL, Nicolau M, et al. Retraction Note: Lysyl oxidase is essential for hypoxia-induced metastasis.Nature · 2020 · 579(7799):456doi:10.1038/s41586-020-2112-4 · PMID 32188947

    The formal retraction notice, published fourteen years after the paper. The indexed text is minimal and does not state the reason, so the date and the fact of retraction are all that can be asserted from PubMed.

16. 16
    Cox TR, Rumney RMH, Schoof EM, et al. The hypoxic cancer secretome induces pre-metastatic bone lesions through lysyl oxidase.Nature · 2015 · 522(7554):106-110doi:10.1038/nature14492 · PMID 26017313

    RETRACTED. As published it reported that lysyl oxidase secreted by low-oxygen breast tumors drives the formation of bone lesions that later host metastases. PubMed carries it as a Retracted Publication, so none of it counts as evidence. It shares its first and last authors with the retracted 2006 Nature paper, which is why the lineage and not only the single paper is the thing to audit.

17. 17
    Cox TR, Rumney RMH, Schoof EM, et al. Retraction Note: The hypoxic cancer secretome induces pre-metastatic bone lesions through lysyl oxidase.Nature · 2023 · 617(7959):208doi:10.1038/s41586-023-06048-x · PMID 37069265

    The formal retraction notice, eight years after the paper and three years after the retraction of the 2006 one. The indexed record carries no abstract, so the date and the fact of retraction are all that can be asserted from PubMed.

18. 18
    Cox TR, Bird D, Baker AM, et al. LOX-mediated collagen crosslinking is responsible for fibrosis-enhanced metastasis.Cancer Res · 2013 · 73(6):1721-32doi:10.1158/0008-5472.CAN-12-2233 · PMID 23345161

    Not retracted, but under an expression of concern since 2019. As published it claimed that lysyl-oxidase-dependent collagen cross-linking creates an environment that permits metastatic colonization and growth, and that cross-linking of type I collagen directly increases tumor cell proliferation. It comes from the same group as the two retracted papers and shares their senior author, so it appears here as part of that audit and not as evidence for anything.

19. 19
    Cox TR, Bird D, Baker AM, et al. Editor's Note: LOX-Mediated Collagen Cross-linking Is Responsible for Fibrosis-Enhanced Metastasis.Cancer Res · 2019 · 79(19):5124doi:10.1158/0008-5472.CAN-19-2419 · PMID 31575632

    The expression of concern on the 2013 paper, indexed by PubMed as such. No abstract is carried, so the existence and the date of the notice are all that can be stated from the record.

20. 20
    Williamson AL, Lustigman S, Oksov Y, et al. Ancylostoma caninum MTP-1, an astacin-like metalloprotease secreted by infective hookworm larvae, is involved in tissue migration.Infect Immun · 2006 · 74(2):961-7doi:10.1128/IAI.74.2.961-967.2006 · PMID 16428741

    The strongest parasite result on this page: a secreted metalloprotease digested collagen, laminin and fibronectin, and antiserum against it cut collagen digestion by 85% and larval migration through tissue by 70-75% against 5-10% for preimmune serum. The authors are developing the same molecule as a vaccine antigen, so a positive result supports their own program.

21. 21
    Thibeaux R, Avé P, Bernier M, et al. The parasite Entamoeba histolytica exploits the activities of human matrix metalloproteinases to invade colonic tissue.Nat Commun · 2014 · 5:5142doi:10.1038/ncomms6142 · PMID 25291063

    One of the two places on this page where a parasite is shown operating the host's own matrix machinery: in human colon explants, the amoeba's own cysteine proteinase A5 activated human pro-MMP-3, which activated pro-MMP-1, and blocking matrix metalloproteinase activity prevented invasion altogether. Human tissue ex vivo; the results are categorical rather than quantified.

22. 22
    Serrat J, López-García M, Torres-Valle M, et al. Early host-parasite interaction models reveal a key role for fibrinolysis in Fasciola hepatica intestinal migration.Parasit Vectors · 2025 · 18(1):387doi:10.1186/s13071-025-06992-9 · PMID 40993705

    The second genuine host-enzyme hijack on this page, and apart from the hookworm antiserum the only parasite result here with both a mechanism and a loss-of-function arm in a living animal: newly excysted juvenile flukes together with host plasminogen generated plasmin in the pericellular space of mouse intestinal epithelial cells, with increased collagen degradation and secretion of urokinase-type plasminogen activator, and inhibiting fibrinolysis pharmacologically reduced juvenile migration in mice. The authors themselves report inter-experimental variability in the animal arm, the cells are mouse rather than human, and the collagen readout is cell culture rather than native tissue.

23. 23
    Robinson MW, Corvo I, Jones PM, et al. Collagenolytic activities of the major secreted cathepsin L peptidases involved in the virulence of the helminth pathogen, Fasciola hepatica.PLoS Negl Trop Dis · 2011 · 5(4):e1012doi:10.1371/journal.pntd.0001012 · PMID 21483711

    The collagen assay the fluke story was said to be missing: FhCL1, FhCL2 and FhCL3 all hydrolyzed native type I and II collagen at neutral pH, but FhCL1 cut only the non-collagenous domains while FhCL2 and FhCL3 cut at multiple sites inside the triple-helical regions of the α1 and α2 chains. It is also the source on this page that names the collagen repeat itself, both as Gly-X-Y and as the Gly-Pro-Xaa motif its peptide substrates were built to mimic. Recombinant enzymes on purified collagen plus molecular simulation; nothing in it tests the enzymes during an infection.

24. 24
    Gelse K, Pöschl E, Aigner T. Collagens--structure, function, and biosynthesis.Adv Drug Deliv Rev · 2003 · 55(12):1531-46doi:10.1016/j.addr.2003.08.002 · PMID 14623400

    A 2003 review of how fibrillar collagens are made and assembled into fibrils, microfibrillar networks and basement membranes. It is also the record of the older count, more than 20 types, which is why the number 28 should be given with its date.

25. 25
    Carnazza M, Quaranto D, DeSouza N, et al. The Duality of Collagens in Metastases of Solid Tumors.Int J Mol Sci · 2025 · 26(19)doi:10.3390/ijms26199745 · PMID 41097013

    A 2025 narrative review that states the dual role explicitly: collagen both enables adhesion, migration, invasion and immune evasion and acts as a structural restraint, and the metastatic cascade begins with local tissue invasion. A review; the dual framing is a synthesis, not a measurement.

26. 26
    Ingram J, Knudsen G, Lim KC, Hansell E, Sakanari J, McKerrow J. Proteomic analysis of human skin treated with larval schistosome peptidases reveals distinct invasion strategies among species of blood flukes.PLoS Negl Trop Dis · 2011 · 5(9):e1337doi:10.1371/journal.pntd.0001337 · PMID 21980548

    The substrate census: on human skin, a cathepsin B2 cut many of the same host proteins as cercarial elastase, so two unrelated enzyme families solve the same problem. It identifies which host skin proteins are cut, which is what makes the unqualified phrase schistosome collagenase an overstatement. It says nothing about pore size, gaps, or whether a larva had to cut anything in order to fit through, so it cannot carry a physical claim about how the organism gets in.

27. 27
    Fishelson Z, Amiri P, Friend DS, et al. Schistosoma mansoni: cell-specific expression and secretion of a serine protease during development of cercariae.Exp Parasitol · 1992 · 75(1):87-98doi:10.1016/0014-4894(92)90124-s · PMID 1639166

    The invasion caught in the act: protease-filled vesicles released into human skin, with ruptured vesicles beside degraded epidermal cells and a degraded dermal-epidermal basement membrane. Ultrastructural and qualitative; no activity was measured against purified collagen.

28. 28
    Provenzano PP, Eliceiri KW, Campbell JM, Inman DR, White JG, Keely PJ. Collagen reorganization at the tumor-stromal interface facilitates local invasion.BMC Med · 2006 · 4(1):38doi:10.1186/1741-7015-4-38 · PMID 17190588

    Where the tumor-associated collagen signatures come from: multiphoton and second-harmonic imaging of intact unfixed mouse mammary tissue defined TACS 1 to 3 and found invasion predominantly along aligned fibers. Descriptive mouse imaging with no effect sizes, confidence intervals or p-values in the abstract.

29. 29
    Zucker S, Cao J, Chen WT. Critical appraisal of the use of matrix metalloproteinase inhibitors in cancer treatment.Oncogene · 2000 · 19(56):6642-50doi:10.1038/sj.onc.1204097 · PMID 11426650

    The contemporary post-mortem on the first wave: phase III trials of marimastat, prinomastat and BAY 12-9566 showed no clinical efficacy and two sponsors discontinued their programs, while the biology turned out to be stromal rather than tumor-cell-autonomous. A review, with no pooled estimates.

30. 30
    Lin X, Hu J, Zhou B, et al. Genotype-phenotype relationship and comparison between eastern and western patients with osteogenesis imperfecta.J Endocrinol Invest · 2024 · 47(1):67-77doi:10.1007/s40618-023-02123-2 · PMID 37270749

    A 671-patient clinical cohort in which COL1A1 and COL1A2 accounted for 55% and 29% of mutations and glycine substitutions produced more severe disease than simply making too little collagen. Observational and single-country; it does not test treatment.

31. 31
    Levental KR, Yu H, Kass L, et al. Matrix crosslinking forces tumor progression by enhancing integrin signaling.Cell · 2009 · 139(5):891-906doi:10.1016/j.cell.2009.10.027 · PMID 19931152

    The experiment behind the stiffness story: inducing collagen cross-linking stiffened the matrix and drove invasion of an oncogene-initiated epithelium, while lowering lysyl-oxidase-mediated cross-linking reduced fibrosis and tumor incidence in MMTV-Neu mice. Animal and in vitro; the abstract gives no effect sizes. It is also not independent of the retracted 2006 paper: both the first author and the senior author of that paper are among this one's thirteen authors.

32. 32
    Chen W, Yang A, Jia J, Popov YV, Schuppan D, You H. Lysyl Oxidase (LOX) Family Members: Rationale and Their Potential as Therapeutic Targets for Liver Fibrosis.Hepatology · 2020 · 72(2):729-741doi:10.1002/hep.31236 · PMID 32176358

    Describes the five copper-dependent lysyl oxidases that cross-link fibrillar collagen and elastin, and states plainly that the individual contribution of each is incompletely understood and that the simtuzumab failures may reflect an inefficient antibody rather than a wrong target. A review arguing for the target class, which is itself a position to note.

33. 33
    Wang K, Ning S, Zhang S, et al. Extracellular matrix stiffness regulates colorectal cancer progression via HSF4.J Exp Clin Cancer Res · 2025 · 44(1):30doi:10.1186/s13046-025-03297-8 · PMID 39881364

    Measures tumor stiffness in 107 patients with magnetic resonance elastography and finds it higher than normal tissue and correlated with collagen content and stage. Observational human imaging plus mechanistic cell work; the human part shows association, not that stiffness caused the stage.

34. 34
    Jiang Z, Xu Y, Wang Y, et al. Matrix stiffness drives squamous cell carcinoma progression via a Piezo1-mediated mechanotransduction feedback loop.J Adv Res · 2025 · 85:935-952doi:10.1016/j.jare.2025.10.041 · PMID 41151627

    Current work measuring real human tumor stiffness by atomic force microscopy — in cutaneous, oral and lung squamous carcinoma against their adjacent normal tissue — and tying it to a mechanical sensor, with a self-reinforcing loop through fibroblasts and collagen deposition. The clinical part is 53 patients and reports a disease-free-survival hazard ratio of 181.03, a figure so extreme it indicates a tiny, unstable estimate rather than a large effect.

35. 35
    Conklin MW, Eickhoff JC, Riching KM, et al. Aligned collagen is a prognostic signature for survival in human breast carcinoma.Am J Pathol · 2011 · 178(3):1221-32doi:10.1016/j.ajpath.2010.11.076 · PMID 21356373

    The first human prognostic test of fiber alignment: in 196 patients, TACS-3 carried hazard ratios between 3.0 and 3.9 for disease-specific and disease-free survival on univariate Cox analysis, with independence from grade, size, estrogen and progesterone receptor status, HER2 status, nodes and subtype reported as a separate claim rather than as the source of those hazard ratios. It is also the paper that states the measurement can be made on routine histopathological sections either by second-harmonic imaging or using picrosirius, which is the cheapest version of the test anyone has proposed. Retrospective, and the abstract reports no confidence intervals and no p-values.

36. 36
    Shoulders MD, Satyshur KA, Forest KT, Raines RT. Stereoelectronic and steric effects in side chains preorganize a protein main chain.Proc Natl Acad Sci U S A · 2010 · 107(2):559-64 (epub 2009)doi:10.1073/pnas.0909592107 · PMID 20080719

    Direct evidence that the three-residue repeat is the whole trick: synthetic (Pro-Pro-Gly) strands form triple helices, and swapping the prolines for 4-fluoroproline or 4-methylproline raised the temperature at which the helix comes apart by more than 50 degrees Celsius. The calorimetry points to an entropic basis for that hyperstability: the substituted side chains restrict the main chain's torsion angles, so the backbone is pre-set into the shape the helix needs before it ever folds. Synthetic chemistry on short peptides, not on human tissue. The print year is 2010 and the electronic publication 31 December 2009, which is why both dates are given.

37. 37
    Mohammad-Rezaei M, Shafaghat Z, Noroozbeygi M, et al. Collagen remodeling in breast cancer progression: from molecular mechanisms to diagnostic and therapeutic opportunities.Clin Transl Oncol · 2026 · online ahead of print (no volume or pages in the PubMed record)doi:10.1007/s12094-026-04450-1 · PMID 42301607

    A date stamp on the current consensus rather than evidence for any single claim: a 2026 review covering the collagen types in breast tissue and how they change, the matrix metalloproteinases and lysyl oxidase, stiffening and aligned fibers, the collagen signatures as candidate biomarkers, second-harmonic imaging and elastography, and enzyme-targeted and nanoparticle approaches. A narrative review with no pooled estimates and no trial data, and its abstract gives no count of the collagen family, so the number 28 on this page still rests on the 2011 review.

38. 38
    Rodrigues J, Karnati M, Raju G, et al. Advancing collagen-related pathology assessment through second-harmonic generation imaging.J Microsc · 2026 · online ahead of print (no volume or pages in the PubMed record)doi:10.1111/jmi.70098 · PMID 42024484

    The current state of the measuring instrument: label-free second-harmonic and polarization-resolved imaging with quantitative descriptors of fiber density, orientation, anisotropy and alignment, increasingly read by machine learning. A methods review that describes promise and reproducibility problems, not a validated clinical test.

39. 39
    Infante E, Castagnino A, Ferrari R, et al. LINC complex-Lis1 interplay controls MT1-MMP matrix digest-on-demand response for confined tumor cell migration.Nat Commun · 2018 · 9(1):2443doi:10.1038/s41467-018-04865-7 · PMID 29934494

    Shows that the cutting is metered: pore size and nuclear stiffness set how much MT1-MMP collagenolysis a cancer cell performs, and the delivery of the enzyme depends on an intact link between nucleus and centrosome. In vitro, with no quantitative effect sizes in the abstract.

40. 40
    Hagedorn EJ, Ziel JW, Morrissey MA, et al. The netrin receptor DCC focuses invadopodia-driven basement membrane transmigration in vivo.J Cell Biol · 2013 · 201(6):903-13doi:10.1083/jcb.201301091 · PMID 23751497

    Live imaging of a single cell crossing a basement membrane in a worm, showing that the invasive protrusion physically displaces basement membrane rather than relying solely on proteolytic dissolution. Caenorhabditis elegans anchor-cell invasion, a developmental model, not a tumor.

41. 41
    Sun X, Wu B, Chiang HC, et al. Tumour DDR1 promotes collagen fibre alignment to instigate immune exclusion.Nature · 2021 · 599(7886):673-678doi:10.1038/s41586-021-04057-2 · PMID 34732895

    Shows that a tumor collagen receptor aligns fibers through its external domain alone, and that neutralizing that domain disrupted alignment and inhibited growth in mouse triple-negative breast cancer. Mouse work; the human part is an uncoefficiented correlation between DDR1 expression and T-cell abundance.

42. 42
    Li C, Sun Y, Shan X, Yang T, Chen G. Matrix stiffness-driven cytoskeletal remodeling and tumor progression in anaplastic thyroid cancer via integrin-focal adhesion kinase signaling.Oncogene · 2026 · 45(6):690-702doi:10.1038/s41388-025-03674-9 · PMID 41606295

    A 2026 engineering-style demonstration that stiffness alone changes behavior: on collagen-coated gels at 60 kilopascals, anaplastic thyroid cancer cells proliferated, migrated and invaded more, and focal adhesion kinase inhibition reversed it. Cells and mice; no patients were treated.

43. 43
    Wolf K, Müller R, Borgmann S, Bröcker EB, Friedl P. Amoeboid shape change and contact guidance: T-lymphocyte crawling through fibrillar collagen is independent of matrix remodeling by MMPs and other proteases.Blood · 2003 · 102(9):3262-9doi:10.1182/blood-2002-12-3791 · PMID 12855577

    The result that broke the assumption that moving through tissue requires cutting it: human T cells crossed 3D collagen with no detectable collagenolysis and were unaffected by a protease inhibitor cocktail, squeezing through existing gaps instead. Leukocytes in gels, not tumor cells in a patient.

44. 44
    Kanatani S, Uhlén P, Barragan A. Infection by Toxoplasma gondii Induces Amoeboid-Like Migration of Dendritic Cells in a Three-Dimensional Collagen Matrix.PLoS One · 2015 · 10(9):e0139104doi:10.1371/journal.pone.0139104 · PMID 26406763

    The counterexample that matters: infected dendritic cells crossed 3D collagen faster in an integrin-independent, amoeboid manner, and blocking five integrins had no significant effect in three dimensions. In vitro; no effect sizes or p-values in the abstract.

45. 45
    Heintzelman MB. Gliding motility in apicomplexan parasites.Semin Cell Dev Biol · 2015 · 46:135-42doi:10.1016/j.semcdb.2015.09.020 · PMID 26428297

    Explains the motor that Toxoplasma and Plasmodium use: myosin anchored to an inner membrane complex drags surface adhesins backward, pushing the parasite forward, with no requirement to cut anything. A review of architecture and mechanism.

46. 46
    Olivares O, Mayers JR, Gouirand V, et al. Collagen-derived proline promotes pancreatic ductal adenocarcinoma cell survival under nutrient limited conditions.Nat Commun · 2017 · 8:16031doi:10.1038/ncomms16031 · PMID 28685754

    The basis for calling collagen a food: pancreatic cancer cells took up collagen fragments, used collagen-derived proline in their metabolism, and needed proline oxidase to proliferate in vitro and in vivo. It does not show labeled collagen being eaten and its proline entering tumor biomass in a living animal.

47. 47
    Davidson SM, Jonas O, Keibler MA, et al. Direct evidence for cancer-cell-autonomous extracellular protein catabolism in pancreatic tumors.Nat Med · 2017 · 23(2):235-241 (epub 2016)doi:10.1038/nm.4256 · PMID 28024083

    Shows that pancreatic tumor cells, and not adjacent normal pancreas, drink and digest extracellular protein in a living mouse, and that blocking that uptake lowers amino acid levels. The protein tracked is albumin, not collagen, which is exactly why the collagen-eating claim remains unfinished.

48. 48
    Dvorák J, Mashiyama ST, Braschi S, et al. Differential use of protease families for invasion by schistosome cercariae.Biochimie · 2008 · 90(2):345-58 (epub 2007)doi:10.1016/j.biochi.2007.08.013 · PMID 17936488

    The reversal of the single-enzyme story: Schistosoma mansoni cercariae secrete serine proteases that could not be identified in S. japonicum, whose glands instead had 40-fold greater cathepsin B-like activity. Biochemical profiling across species; it does not test which enzyme is sufficient in a host.

49. 49
    McKerrow JH, Jones P, Sage H, Pino-Heiss S. Proteinases from invasive larvae of the trematode parasite Schistosoma mansoni degrade connective-tissue and basement-membrane macromolecules.Biochem J · 1985 · 231(1):47-51doi:10.1042/bj2310047 · PMID 3904737

    The one purified schistosome enzyme that was actually put on collagen: the proteinase purified from S. mansoni cercariae degraded fibronectin and laminin and behaved as a type-specific collagenase against the basement-membrane collagens IV and VIII, but not against the interstitial collagens I, III and V. Forty years old, measured on labeled matrices made by cultured cells rather than on human skin, and a purification rather than a loss-of-function test — but it is why the flat claim that no schistosome enzyme has been shown to cut collagen is wrong for type IV.

50. 50
    McKerrow JH, Pino-Heiss S, Lindquist R, Werb Z. Purification and characterization of an elastinolytic proteinase secreted by cercariae of Schistosoma mansoni.J Biol Chem · 1985 · 260(6):3703-7PMID 3882713

    The companion purification of cercarial elastase itself: a serine proteinase of molecular weight 30,000 with a pH optimum of 9, which degrades elastin, Azocoll, gelatin, laminin, fibronectin, keratin and type IV collagen. The PubMed record carries no DOI. Again a substrate list from a purified enzyme, with no infection experiment and no interstitial collagen cut.

51. 51
    Corvo I, Cancela M, Cappetta M, Pi-Denis N, Tort JF, Roche L. The major cathepsin L secreted by the invasive juvenile Fasciola hepatica prefers proline in the S2 subsite and can cleave collagen.Mol Biochem Parasitol · 2009 · 167(1):41-7doi:10.1016/j.molbiopara.2009.04.005 · PMID 19383516

    The earlier of the two fluke collagen papers: FhCL3, expressed in yeast, showed a 70-fold preference for a proline-containing substrate over typical cathepsin substrates and cleaved type I collagen efficiently across a range of pH and temperature, while not cleaving immunoglobulin. The authors place it beside human cathepsin K and plant zingipain as one of the few cysteine proteases that can cut collagen at all. Recombinant protein in a tube; the step from that to tissue invasion is their suggestion and is not tested.

52. 52
    Muñoz ML, Calderón J, Rojkind M. The collagenase of Entamoeba histolytica.J Exp Med · 1982 · 155(1):42-51doi:10.1084/jem.155.1.42 · PMID 6274995

    The actual first report of amoebic collagenolysis, ten years before the collagen-binding-protein paper: trophozoites on reconstituted type I collagen gels produced a defined area of lysis and digested three and a half times more labeled type I than type III collagen, with the activity at the plasma membrane and direct contact with the substrate required. Quantified in vitro; no enzyme was ever identified as a protein here.

53. 53
    Gadasi H, Kessler E. Correlation of virulence and collagenolytic activity in Entamoeba histolytica.Infect Immun · 1983 · 39(2):528-31doi:10.1128/iai.39.2.528-531.1983 · PMID 6299942

    The independent replication, from a different group a year later: media and extracts of the highly virulent HM-1 strain degraded native type I collagen while the less virulent 200-NIH and HK-9 strains showed significantly lower activity, with the activity again at the plasma membrane. Three strains and a gel readout, so the correlation with virulence rests on very few comparisons.

54. 54
    Hotez P, Haggerty J, Hawdon J, et al. Metalloproteases of infective Ancylostoma hookworm larvae and their possible functions in tissue invasion and ecdysis.Infect Immun · 1990 · 58(12):3883-92doi:10.1128/iai.58.12.3883-3892.1990 · PMID 2254016

    The old comparison that long carried the weight for Strongyloides: hookworm larvae degraded fibronectin but not solubilized elastin or laminin, while Strongyloides stercoralis degraded all three. In vitro substrate gels and a skin chamber; it names no enzyme itself, but the Strongyloides enzyme was named, sized and cloned in other papers, and a collagen result for the genus exists from 1987.

55. 55
    Rosales-Encina JL, Campos-Salazar MS, Rojkind Matluk M. Entamoeba histolytica collagen binding proteins.Arch Med Res · 1992 · 23(2):109-13PMID 1340269

    The original claim that the amoeba carries its own collagen-binding proteins at 105, 56 and 30 kilodaltons with collagenolytic activity resembling a mammalian collagenase. The 30-kilodalton receptor was assigned on the authors' own indirect evidence in 1992, with no modern confirmation located and no DOI in the record. The collagenolytic activity around that claim is not the unconfirmed part: it was reported in 1982 and reproduced by an independent group in 1983. It is the 30-kilodalton receptor that has never been confirmed.

56. 56
    Muñoz ML, Lamoyi E, León G, et al. Antigens in electron-dense granules from Entamoeba histolytica as possible markers for pathogenicity.J Clin Microbiol · 1990 · 28(11):2418-24doi:10.1128/jcm.28.11.2418-2424.1990 · PMID 2174899

    Where the amoebic activity was finally given a number and a place: contact with collagen induced the formation and release of electron-dense granules, and purified granules contained 1.66 units of collagenase per milligram of protein. The paper's own purpose is a diagnostic antibody, and the collagenase figure is a single measurement on one purified fraction.

57. 57
    Provenzano PP, Inman DR, Eliceiri KW, et al. Collagen density promotes mammary tumor initiation and progression.BMC Med · 2008 · 6:11doi:10.1186/1741-7015-6-11 · PMID 18442412

    The causal animal experiment: in a bi-transgenic mouse with increased mammary stromal collagen, tumor formation rose about threefold (p < 0.00001) and lung metastases about threefold (p < 0.05). One engineered mouse model; p-values are given without effect-size confidence intervals.

58. 58
    O'Kelly E, Cwiklinski K, De Marco Verissimo C, et al. Moonlighting on the Fasciola hepatica tegument: Enolase, a glycolytic enzyme, interacts with the extracellular matrix and fibrinolytic system of the host.PLoS Negl Trop Dis · 2024 · 18(8):e0012069doi:10.1371/journal.pntd.0012069 · PMID 39213442

    How the fluke gets hold of host plasminogen in the first place: a glycolytic enzyme that also sits on the parasite's surface and in its secretions binds laminin, binds plasminogen, and speeds its conversion to plasmin in the presence of either host activator. Binding assays on recombinant protein; it bound laminin but not fibronectin, and collagen binding was not tested.

59. 59
    Ramaswamy A, Bhargava P, Gota V, et al. Efficacy of losartan plus modified FOLFIRINOX versus modified FOLFIRINOX in advanced pancreatic cancers: A randomized clinical trial (AFPAC Study).Cancer · 2025 · 131(13):e35945doi:10.1002/cncr.35945 · PMID 40542713

    A third randomized attempt on a tumor's connective tissue, by a different route: 88 patients with advanced pancreatic cancer randomized to chemotherapy with or without losartan, a blood-pressure drug used here to damp the signals that build stroma. Overall survival hazard ratio 0.76 (95% CI 0.47-1.22, p = .392), median 10.4 against 9.1 months, no difference in response rate and no difference in the plasma TGF-β trend, and the trial will not proceed to full accrual. This was an unplanned early analysis looking for a signal, so it is a trial stopped for futility rather than a definitive negative one.

60. 60
    Vachhani P, Tan P, Watson AM, et al. A phase I/IIa trial of PXS-5505, a novel pan-lysyl oxidase inhibitor, in advanced myelofibrosis.Haematologica · 2025 · 110(10):2376-2387doi:10.3324/haematol.2024.287231 · PMID 40241543

    The one place the cross-linking idea has reached patients without simply failing: 24 patients of median age 72 with relapsed or refractory myelofibrosis, a neoplasm of blood-forming cells with scarring of the marrow, took a small-molecule inhibitor of all the lysyl oxidases; 200 mg twice daily was tolerated, systemic lysyl oxidase activity fell robustly, 13 of 24 completed 24 weeks, and the authors report preliminary indications of effect including reduced marrow collagen. Single-arm and uncontrolled by design, so none of that is a demonstrated outcome benefit, and nine of the authors are employees of Syntara, the developer.

61. 61
    Tian Y, Li Q. LOXL4: a key regulatory factor at the intersection of fibrosis and tumor progression.Hum Cell · 2026 · 39(10)doi:10.1007/s13577-026-01452-7 · PMID 42749970

    A current review of one lysyl oxidase family member, useful mainly for its own admission that clinical development of LOXL4-targeted therapy remains limited by functional heterogeneity, missing biomarkers and compensation within the family. A narrative review with no pooled estimates.

62. 62
    Udin MH, Nepali S, Sonkawade S, et al. Tumor-associated collagen signature in relation to metastasis in lung adenocarcinoma.J Thorac Dis · 2026 · 18(6):653doi:10.21037/jtd-2026-1-0356 · PMID 42444926

    The architecture-not-amount result, and the first extension of the collagen signature idea beyond the breast: in 120 lung adenocarcinomas, 54 with nodal metastasis and 66 without, fiber alignment measured 0.87 against 0.78 (P=0.003), elastin volume fraction 0.29 against 0.38 (P=0.001), the elastin-to-collagen ratio fell (P=0.002), and total collagen content did not differ at all (P=0.87). Retrospective, from a single center, with specimens from 2005 to 2022, and the endpoint is nodal status rather than survival, so it is weaker than the breast datasets even though the null on total collagen is the most useful number in it.

63. 63
    Wang S, Su J, Han X, et al. From Label-Free Multiphoton Imaging to Pathological Reports: A Vision-Language Breast Cancer Margin Pathological Diagnosis System.Adv Sci (Weinh) · 2026 · 13(45):e75709doi:10.1002/advs.75709 · PMID 42136540

    The concrete machine-learning instance this page's claim about machine learning needed: a model that reads collagen signatures from a single unstained multiphoton section of a surgical margin and returns a virtual stained image, a pixel-level margin heatmap and a written description, validated on 158 invasive breast cancer specimens. It is benchmarked against other pathology vision-language models rather than against patient outcomes, and it shares authors with the 995-patient and 941-patient signature studies, so it is not an independent assessment of the signatures it reads.

64. 64
    Reitsam NG, Märkl B, Nagtegaal ID, Svrcek M. Direct tumour-adipocyte contact / Stroma AReactive Invasion Front Area (TAC/SARIFA) predicts adverse outcome in colorectal cancer: a focused evidence synthesis and meta-analysis.Histopathology · 2026 · 89(5):835-848doi:10.1111/his.70237 · PMID 42460779

    The counterweight to the collagen-is-bad story, and more specific than it is usually reported to be: what is scored is direct tumor-adipocyte contact at the invasion front, with no intervening stroma or inflammatory infiltrate, and not an absence of collagen. Pooled across 7 cohorts and 5,932 patients it carried a cancer-specific survival hazard ratio of 1.77 (95% CI 1.55-2.01), and across 5 cohorts and 4,545 patients an overall survival hazard ratio of 1.53 (1.39-1.68), from literature searched to 7 May 2026. All included studies were retrospective, and the authors invented the marker they are assessing.

65. 65
    Shang A, Yu P, Li L, He G, Xu J. Tumor-stroma ratio as a clinical prognostic factor in colorectal carcinoma: A meta-analysis of 7,934 patients.Oncol Lett · 2025 · 29(4):190doi:10.3892/ol.2025.14936 · PMID 40041409

    Colorectal evidence that points the same way as the breast evidence rather than against it: pooled across 21 studies published between 2007 and 2023 and 7,934 patients, a high tumor-stroma ratio carried hazard ratios of 1.84 (95% CI 1.44-2.34) for overall survival, 1.85 (1.27-2.68) for disease-free survival and 1.97 (1.46-2.65) for cancer-specific survival, with no association in stage I disease. The abstract does not state which way round the ratio is scored, so it is cited for the existence and size of an association with stromal quantity, not for a threshold.

66. 66
    Liu ZJ, Zhang XW, Liu QQ, Wang SZ. Correlation analysis of interstitial maturity and prognosis of colorectal cancer: Meta-analysis.World J Gastrointest Oncol · 2024 · 16(6):2816-2825doi:10.4251/wjgo.v16.i6.2816 · PMID 38994137

    The other colorectal direction, scored as the character of the stroma rather than its amount: across 19 cohorts in 15 papers and 9,849 patients, immature stroma against mature carried hazard ratios of 2.66 (95% CI 2.30-3.08) for relapse-free survival, 3.68 (2.33-5.81) for disease-free survival and 1.70 (1.53-1.87) for overall survival. All component studies are observational, the groupings are a pathologist's categories, and the reporting in the abstract is uneven.

67. 67
    Labernadie A, Kato T, Brugués A, et al. A mechanically active heterotypic E-cadherin/N-cadherin adhesion enables fibroblasts to drive cancer cell invasion.Nat Cell Biol · 2017 · 19(3):224-237doi:10.1038/ncb3478 · PMID 28218910

    Shows fibroblasts physically towing cancer cells through matrix by a direct N-cadherin to E-cadherin junction that strengthens under force, with the junctions also seen in patient-derived material. Forces were measured with optical tweezers but no values appear in the abstract.

68. 68
    Shoari A, Coban MA, Radisky ES. Therapeutic potential of inhibitors of gelatinases MMP-2 and MMP-9 for the treatment of breast cancer.FEBS J · 2025 · 293(13):4014-4042doi:10.1111/febs.70343 · PMID 41313679

    A current account of the two enzymes specialized for cutting type IV collagen, and of the selective inhibitors, antibodies, nanobodies and engineered inhibitors now being built after the failure of broad-spectrum drugs. A review of preclinical strategies; it reports no completed trial of a selective agent, although one selective agent, an antibody against MMP-9, had in fact completed a randomized phase III four years earlier with a negative result, which this page cites directly.

69. 69
    Ghose R, Sanz-Moreno V. Amoeboid cancer cells at a glance.J Cell Sci · 2026 · 139(10):jcs264674doi:10.1242/jcs.264674 · PMID 42212523

    The current statement of the amoeboid cell state and of what distinguishes it from the mesenchymal one: loss of cell-cell junctions, a rounded low-adhesion shape, dependence on Rho-ROCK-myosin II cortical contractility, blebs used as protrusions, enrichment at invasive fronts and in therapy-resistant populations, and induction by confinement, matrix texture and shear. A review and teaching poster, so it is a synthesis with no effect sizes and no patient data.

70. 70
    Cumming J, Maneshi P, Dongre M, et al. Dissecting FAP+ Cell Diversity in Pancreatic Cancer Uncovers an Interferon-Response Subtype of Cancer-Associated Fibroblasts with Tumor-Restraining Properties.Cancer Res · 2025 · 85(13):2388-2411doi:10.1158/0008-5472.CAN-23-3252 · PMID 40215177

    Why an argument about fibroblasts as a drug target has to say which fibroblasts: single-cell profiling of the matrix-building cells in human pancreatic cancer resolved inflammatory and myofibroblastic subtypes and uncovered a third, interferon-response subtype, and inducing that state suppressed tumor cell invasiveness and turned tumor-associated neutrophils against the tumor. Human tissue for the profiling, then organoid coculture and mice for the induction; the tumor-restraining effect is a model result, not a patient outcome.

71. 71
    De Giuseppe A, Deichsel A, Reese A, et al. Collagen-binding integrin α11β1 contributes to joint destruction in arthritic hTNFtg mice.Ann Rheum Dis · 2025 · 84(10):1649-1659doi:10.1016/j.ard.2025.07.011 · PMID 40816940

    The non-cancer control case for the same machinery: rheumatoid synovial fibroblasts behave invasively, and deleting the collagen-binding integrin α11β1 reduced cartilage degradation and bone erosion in arthritic mice. The authors advance α11 as a therapeutic target, which is an intellectual stake to note.

72. 72
    Legendre C, Dayan G, Rousselle P. When proteases reshape barriers: Basement membrane remodelling in development, wound healing and tumour progression.FEBS J · 2026 · 293(13):3954-3972doi:10.1111/febs.70585 · PMID 42157512

    The current account of basement-membrane remodeling, and the best single source for the point that cancer uses the same toolkit as development and wound repair: the matrix metalloproteinases, the membrane-tethered MT-MMPs, the ADAMTS proteases and non-proteolytic regulators thin, perforate and stiffen these sheets in patterned, transient ways during normal morphogenesis and healing, while in carcinomas invadopodia concentrate the same enzymes to perforate them and the fragments left behind — collagen IV escape tracks and cut laminin pieces — act as signals for proliferation, survival and new blood vessels. A narrative review with no pooled estimates and no clinical data.

73. 73
    McGwire BS, Chang KP, Engman DM. Migration through the extracellular matrix by the parasitic protozoan Leishmania is enhanced by surface metalloprotease gp63.Infect Immun · 2003 · 71(2):1008-10doi:10.1128/IAI.71.2.1008-1010.2003 · PMID 12540585

    A surface metalloprotease that degrades type IV collagen and fibronectin and whose overexpression improved migration through matrix. A three-page report with no magnitudes in the abstract, and overexpression is not the same as showing the enzyme is required.

74. 74
    Li Z, Kang D, Wang C, et al. Prognostic value of nuclear features based on tumor-associated collagen signatures in breast cancer.NPJ Breast Cancer · 2025 · 11(1):148doi:10.1038/s41523-025-00860-6 · PMID 41392087

    The most recent human extension, moving the measurement onto ordinary stained slides: nuclear features defined relative to collagen signatures in 941 patients were an independent prognostic factor. The abstract reports no hazard ratios, AUCs or confidence intervals, so the size of the gain cannot be quoted.

75. 75
    Zhao X, Fan Y, Yang J, Gao H. Single-cell transcriptomics identify mechanical-memory-associated cell states in metastatic HR breast cancer.Breast Cancer Res · 2026 · 28(1)doi:10.1186/s13058-026-02280-1 · PMID 41998694

    A 2026 attempt to show that cells remember past stiffness: a 79-gene signature was elevated in aggressive tumors, and stiffening 3D collagen raised RELA and MMP14 expression. Computational signature plus cell culture; the human component is survival association in public datasets, not a prospective test.

76. 76
    Ingram JR, Rafi SB, Eroy-Reveles AA, et al. Investigation of the proteolytic functions of an expanded cercarial elastase gene family in Schistosoma mansoni.PLoS Negl Trop Dis · 2012 · 6(4):e1589doi:10.1371/journal.pntd.0001589 · PMID 22509414

    Establishes that the invasion enzyme is a whole expanded gene family unique to this species, likely a matter of gene dosage, and that it is switched on before the larva even leaves the snail. Genomic, proteomic and activity profiling, with substrate differences predicted by modeling rather than measured on collagen.

77. 77
    Chlichlia K, Schauwienold B, Kirsten C, Doenhoff MJ, Fishelson Z, Ruppel A. Schistosoma japonicum reveals distinct reactivity with antisera directed to proteases mediating host infection and invasion by cercariae of S. mansoni or S. haematobium.Parasite Immunol · 2005 · 27(3):97-102doi:10.1111/j.1365-3024.2005.00751.x · PMID 15882236

    Immunological mapping showing that one elastase isoform is detectable in S. mansoni and S. haematobium but apparently absent from the glands of S. japonicum. Antibody localization; absence of staining is weaker than absence of the gene.

78. 78
    Santana JM, Grellier P, Schrével J, Teixeira AR. A Trypanosoma cruzi-secreted 80 kDa proteinase with specificity for human collagen types I and IV.Biochem J · 1997 · 325(Pt 1):129-37doi:10.1042/bj3250129 · PMID 9224638

    The primary paper the T. cruzi collagen claim turned out to have: an 80-kilodalton proteinase, purified from the parasite forms that infect mammalian cells and secreted by them, hydrolyzed purified human collagen types I and IV at neutral pH while leaving albumin, laminin, immunoglobulin G, insulin and cytochrome c alone, and cut native type I collagen in rat mesentery. A purified enzyme with a clean specificity; the step from that to facilitating infection is the authors' suggestion, and they also propose it as a drug target.

79. 79
    Jílková A, Rubešová P, Fanfrlík J, et al. Druggable Hot Spots in the Schistosomiasis Cathepsin B1 Target Identified by Functional and Binding Mode Analysis of Potent Vinyl Sulfone Inhibitors.ACS Infect Dis · 2021 · 7(5):1077-1088 (epub 2020)doi:10.1021/acsinfecdis.0c00501 · PMID 33175511

    Important for keeping two jobs apart: this schistosome cathepsin B1 is described as a peptidase for digesting host blood protein and a validated drug target, with subnanomolar vinyl sulfone inhibitors and two crystal structures. Nutrition and drug discovery, not invasion.

80. 80
    Orrico-Ferreira CM, Cruz PEO, Rios JVB, et al. Getting into host's skin: initial immune response to Schistosoma mansoni infection.Front Immunol · 2025 · 16:1661465doi:10.3389/fimmu.2025.1661465 · PMID 41479896

    Cited for the second job of cercarial elastase, which is better supported than its collagen-cutting one: the enzyme interferes with how host cells present parasite antigens and with antibody function, making it an immune-evasion molecule at the skin interface, alongside the parasite-driven shift from a Th1 to a Th2 response. A review written from an immunology standpoint; it carries no collagen or matrix measurements. The PubMed record drops the species name from the title.

81. 81
    Zhang T, Mo XJ, Xu B, et al. Enzyme activity of Schistosoma japonicum cercarial elastase SjCE-2b ascertained by in vitro refolded recombinant protein.Acta Trop · 2018 · 187:15-22doi:10.1016/j.actatropica.2018.07.009 · PMID 30017495

    Measures the enzyme directly after refolding it from inclusion bodies: a Michaelis constant of 0.116 millimolar — the concentration at which the enzyme works at half its top speed, and a symbol the PubMed record garbles — and a specific activity of 1,900 nanomoles of p-nitroaniline per minute per milligram, both measured on a synthetic peptide substrate. Recombinant protein in a tube, not collagen and not skin.

82. 82
    Dolecková K, Kasný M, Mikes L, et al. The functional expression and characterisation of a cysteine peptidase from the invasive stage of the neuropathogenic schistosome Trichobilharzia regenti.Int J Parasitol · 2009 · 39(2):201-11 (epub 2008)doi:10.1016/j.ijpara.2008.06.010 · PMID 18708063

    A bird schistosome's cathepsin B, expressed in yeast, degraded skin, serum and nervous tissue proteins; its peptidolytic activity was characterized on synthetic fluorogenic substrates at an optimal pH of 6.0, and it separately showed elastinolytic activity, which the authors propose substitutes for the serine elastase other species use. Recombinant enzyme on tissue extracts; no collagen assay and no infection experiment.

83. 83
    González-Rivas E, González-Jiménez D, De León Rendón JL, et al. Host-parasite interactions during Entamoeba histolytica infection: an ex vivo human intestinal model.Front Cell Infect Microbiol · 2026 · 16:1870104doi:10.3389/fcimb.2026.1870104 · PMID 42553089

    The current human-tissue model of amoebic invasion, a decade on from the colon explants: thin living slices of human intestine stayed viable for 24 hours and showed progressive epithelial damage and invasion into the connective tissue beneath, while the parasite's adhesion lectin held a steady expression level and its cysteine protease genes switched on early and its stress-response genes rose later, with cytokine kinetics that do not fit a simple Th1-to-Th2 switch. Gene expression and histology only: no collagen measurement, no matrix readout, and no quantification of invasion depth. The PubMed record drops the species name from the title.

84. 84
    Watanabe Costa R, da Silveira JF, Bahia D. Interactions between Trypanosoma cruzi Secreted Proteins and Host Cell Signaling Pathways.Front Microbiol · 2016 · 7:388doi:10.3389/fmicb.2016.00388 · PMID 27065960

    A review that asserts Trypanosoma cruzi secretes factors producing proteolytic degradation of host collagens, and gives 6 to 7 million infected people in Latin America. The collagen claim reads as a review assertion, but a primary paper does stand behind it after all: a 1997 purification of a secreted 80-kilodalton proteinase with activity on human collagen types I and IV, which this page now cites directly instead of relying on this review.

85. 85
    Farhab M, Yuan YG. From nucleation to capping: The lifecycle of an actin filament in Toxoplasma gondii gliding.Vet Parasitol · 2025 · 342:110679doi:10.1016/j.vetpar.2025.110679 · PMID 41455416

    The current account of why that motor is unusual: the parasite's actin is mostly monomeric and turns over very fast, and myosin A walks on short dynamic filaments to translocate adhesins. A review aimed at drug and vaccine targets.

86. 86
    Kumar A, Vadas O, Dos Santos Pacheco N, et al. Structural and regulatory insights into the glideosome-associated connector from Toxoplasma gondii.Elife · 2023 · 12:e86049doi:10.7554/eLife.86049 · PMID 37014051

    The structural core of gliding: a crystal structure of the conserved protein that couples actin filaments to the surface adhesins and carries the motor's force to the ground, showing a supercoiled armadillo-repeat arm that closes into a ring, with solution and binding data suggesting it also opens into extended forms. Structural biology on an isolated protein; the multi-conformational model of how it works inside the glideosome is proposed, not shown in a moving parasite. The PubMed record drops the organism from the title.

87. 87
    Dos Santos Pacheco N, Brusini L, Haase R, et al. Conoid extrusion regulates glideosome assembly to control motility and invasion in Apicomplexa.Nat Microbiol · 2022 · 7(11):1777-1790doi:10.1038/s41564-022-01212-x · PMID 36109645

    Shows the motor is gated rather than standing ready: expansion microscopy and genetics in Toxoplasma gondii established that the preconoidal rings at the parasite's tip serve as a hub for glideosome components including Formin1, identified two conserved ring proteins essential for the structure, and showed that extruding the conoid directs the flux of actin filaments to the surface and so controls motility and invasion. Genetics in one organism; the components are described as conserved across the group but were not tested across it.

88. 88
    Douglas RG, Reinig M, Neale M, Frischknecht F. Screening for potential prophylactics targeting sporozoite motility through the skin.Malar J · 2018 · 17(1):319doi:10.1186/s12936-018-2469-0 · PMID 30170589

    States the premise plainly: migration of sporozoites through the dermis to find a blood vessel is the first obligatory step of human infection, and a screen for compounds that stop that movement found very few, with a topical gel failing in mice. Preclinical screening.

89. 89
    Formaglio P, Wosniack ME, Tromer RM, et al. Plasmodium sporozoite search strategy to locate hotspots of blood vessel invasion.Nat Commun · 2023 · 14(1):2965doi:10.1038/s41467-023-38706-z · PMID 37221182

    Better evidence than an engineered surrogate that malaria sporozoites navigate rather than digest: imaging them inside living skin showed a fast, far-ranging search pattern of the kind that is mathematically efficient for finding scarce targets, switching on contact with a vessel to a slow local probing mode concentrated at entry sites marked by pericytes. A rodent malaria model, and the characterization of the search pattern is a statistical inference from the shape of the tracks rather than a mechanism.

90. 90
    Muthinja MJ, Ripp J, Hellmann JK, et al. Microstructured Blood Vessel Surrogates Reveal Structural Tropism of Motile Malaria Parasites.Adv Healthc Mater · 2017 · 6(6)doi:10.1002/adhm.201601178 · PMID 28117558

    Shows malaria sporozoites navigating by shape rather than by digestion: in pillar arrays the crescent-shaped parasites preferentially associated with and circled pillars matching their own curvature. An engineered surrogate, not real dermis; imaging the same parasites inside living skin has since answered the navigation question better.

91. 91
    Cancela M, Acosta D, Rinaldi G, et al. A distinctive repertoire of cathepsins is expressed by juvenile invasive Fasciola hepatica.Biochimie · 2008 · 90(10):1461-75doi:10.1016/j.biochi.2008.04.020 · PMID 18573308

    The invading juvenile liver fluke makes its own cathepsin set, and prefers substrates with proline in the P2 position, a preference otherwise shared with vertebrate cathepsin K. That is a hint of collagen-cutting specificity rather than a collagen assay, and no collagen was tested here — but the assay it pointed to was done by other groups in 2009 and 2011, and the enzymes do cut collagen.

92. 92
    Hou WS, Li Z, Gordon RE, et al. Cathepsin K is a critical protease in synovial fibroblast-mediated collagen degradation.Am J Pathol · 2001 · 159(6):2167-77doi:10.1016/S0002-9440(10)63068-4 · PMID 11733367

    The human comparison enzyme for parasite cathepsins: cathepsin K, the collagen-cutting cysteine protease, degrades collagen inside fibroblast lysosomes after fibrils are taken up, and inhibitors of cathepsins L, B and S do not substitute. Cell and tissue work in rheumatoid arthritis, not cancer.

93. 93
    Gordon CA, Utzinger J, Muhi S, et al. Strongyloidiasis.Nat Rev Dis Primers · 2024 · 10(1):6doi:10.1038/s41572-023-00490-x · PMID 38272922

    The current clinical reference for the worm that penetrates skin directly, reproduces inside the host and can reinfect it, becoming fatal in immunosuppressed people. A disease primer; it carries no enzymology and states that global prevalence is largely unknown.

94. 94
    McKerrow JH, Brindley P, Brown M, Gam AA, Staunton C, Neva FA. Strongyloides stercoralis: identification of a protease that facilitates penetration of skin by the infective larvae.Exp Parasitol · 1990 · 70(2):134-43doi:10.1016/0014-4894(90)90094-s · PMID 2137091

    The paper that names the Strongyloides invasion enzyme and gives it a function: infective larvae secrete a histolytic metalloprotease with elastase activity that degrades a model of dermal extracellular matrix, and metalloprotease inhibitors prevented larvae from invading skin in vitro. It also reports that the larvae can move through tissue at up to 10 centimeters per hour. Skin invasion in vitro rather than in an animal, and no assay on purified collagen.

95. 95
    Brindley PJ, Gam AA, McKerrow JH, Neva FA. Ss40: the zinc endopeptidase secreted by infective larvae of Strongyloides stercoralis.Exp Parasitol · 1995 · 80(1):1-7doi:10.1006/expr.1995.1001 · PMID 7529715

    Sizes and names the same activity: a zinc endopeptidase of 40 kilodaltons and isoelectric point 5, with broad specificity for constituents of dermal matrix including elastin, immunogenic in infected people and able to release histamine from their leukocytes. Zymogram and immunoprecipitation work; the skin-penetration claim is carried over from the 1990 inhibitor experiment rather than repeated here.

96. 96
    Gomez Gallego S, Loukas A, Slade RW, et al. Identification of an astacin-like metallo-proteinase transcript from the infective larvae of Strongyloides stercoralis.Parasitol Int · 2005 · 54(2):123-33doi:10.1016/j.parint.2005.02.002 · PMID 15866474

    Clones the likely molecule behind Ss40: strongylastacin, a zinc metalloendopeptidase of the metzincin superfamily carrying the astacin active-site motifs, transcribed only in the infective third-stage larva and structurally related to crayfish astacin and to the hookworm enzyme family. A sequence and expression study; the authors call it the probable candidate for Ss40, which is an inference, and no enzyme was purified or assayed here.

97. 97
    Rege AA, Dresden MH. Strongyloides spp.: demonstration and partial characterization of acidic collagenolytic activity from infective larvae.Exp Parasitol · 1987 · 64(3):275-80doi:10.1016/0014-4894(87)90037-3 · PMID 2824230

    The only collagen result for this genus: infective larvae of Strongyloides ransomi, S. ratti and S. stercoralis all showed collagenolytic activity in acidic homogenates, resembling vertebrate acidic cysteine proteinases, after attempts to find a classical neutral-pH collagenase in S. ratti had failed. Homogenates at acidic pH, not a secreted enzyme at the pH of skin, which is why it does not settle the question.

98. 98
    Koseki K, Yamamoto A, Tanimoto K, et al. Dityrosine Crosslinking of Collagen and Amyloid-β Peptides Is Formed by Vitamin B12 Deficiency-Generated Oxidative Stress in Caenorhabditis elegans.Int J Mol Sci · 2021 · 22(23):12959doi:10.3390/ijms222312959 · PMID 34884761

    Cited only for the nematode side of collagen chemistry: worm cuticle collagen matures outside the cell by dityrosine cross-linking rather than by a lysyl oxidase, and the worm's prolyl 4-hydroxylase uses ascorbate as a coenzyme just as the human enzyme does, so a deficiency that stripped the worm of ascorbate lowered its collagen and raised its dityrosine cross-linking. Caenorhabditis elegans is free-living and infects nobody, and the paper's own subject is vitamin B12 deficiency and amyloid, so nothing here reaches a parasitic worm's cuticle without assuming it. The abstract's wording for the size of the fall in collagen is ambiguous, so no percentage is quoted from it, and the PubMed record drops the vitamin's number and the organism's name from the title.

99. 99
    Naowaratwattana W, Dokduang H, Daiponmak W, et al. Periductal Fibrosis and Cholangiocarcinoma-Related Outcomes in Liver Fluke-Endemic Regions: A Systematic Review and Meta-Analysis.Med Sci (Basel) · 2026 · 14(3):380doi:10.3390/medsci14030380 · PMID 42506349

    The closest thing anyone has to a measurement of parasitized tissue architecture followed to a cancer endpoint in people: in regions endemic for the liver fluke Opisthorchis viverrini, periductal fibrosis seen on ultrasound carried a pooled odds ratio of 2.77 (95% CI 2.24-3.44, with no heterogeneity between studies) for cholangiocarcinoma-related outcomes across 3 studies and 758,686 participants, with 4 studies and 759,117 participants contributing prevalence. Only five studies met inclusion, all observational, outcome definitions and adjustment differed between them, and the authors say in their own abstract that this is a summary association and that ultrasound fibrosis is neither a histological diagnosis nor a proven causal precursor.