Which ingredients occlude dentine tubules, and how well? A ranked review of the SEM evidence.

Rank the published electron-microscope evidence on its own terms and hydroxyapatite comes out at the top of it: in the one in vitro study that set the three main occluding actives side by side, a 15% nano-hydroxyapatite paste occluded about 98% of the tubules on human dentine discs after seven days, against about 83% for calcium sodium phosphosilicate and about 69% for arginine¹. Then read the other measurements printed inside those same papers — fluid flow, hardness, elemental analysis, what survives an acid drink — and the ranking stops predicting anything a person would feel. That is not an argument against occlusion. It is the reason a sealed tubule is where the evidence starts rather than where it finishes, and it is why this page ranks the ingredients first and then spends longer on what the ranking cannot tell you. S3 Sensitivity Science™ prints 10% nano-hydroxyapatite and 5% biomimetic hydroxyapatite on its own pack as inclusion levels of the ingredient as supplied, and says in the same breath that the active hydroxyapatite content is lower — which is exactly the kind of distinction this page spends its length making about everybody else's numbers.

What was checked24 peer-reviewed studies and guidance from the Oral Health Foundation, plus product information as published by each brand

What does "seals dentine tubules" mean under a microscope?

It means a photograph, taken at magnifications around two thousand times, of a slice of dentine that somebody cut, etched with acid to open the tubules, brushed with a paste, and then dried, coated and put in a vacuum. The Oral Health Foundation puts the mechanism in one sentence — sensitive toothpastes work by blocking the tiny channels in dentine, and the effect is maintained only while you keep using them²⁵. Everything below is what happens when you ask that sentence for its evidence. The anatomy itself — what a tubule is, how many of them there are, where they run — is covered in the Journal's piece on what are dentine tubules, and this page goes straight past it to the measurements.

Three different things get called "occlusion" in these papers, and they are not interchangeable.

The first is a count: the proportion of tubule openings in a field of view that are no longer open. That is where 98.1%, 83.1% and 69.1% come from¹, and where the 66.13% recorded for a 15% nano-hydroxyapatite paste in a second in vitro count comes from, in an experiment whose test paste was donated by the maker of the hydroxyapatite it contained².

The second is a judgement on a scale. One in vitro study of three desensitising pastes scored each specimen from one to five, with a lower number meaning more occlusion, and reported means of 2.20, 2.30, 3.60 and 5.00 for the water control³. Another scored its specimens on a five-category scale running the other way — one for occluded, five for unoccluded, with three named categories in between — read by three calibrated, blinded evaluators¹². Neither of those numbers converts into the other, and neither converts into a percentage.

The third is a depth. Serial block-face electron microscopy cuts the specimen away layer by layer, so it can say how far into the channel a deposit reached rather than only whether the mouth of the channel is covered; in that in vitro model a calcium sodium phosphosilicate paste blocked 100% of the surface tubules against 83% for a stannous fluoride paste, and the treated dentine's greyscale value, a proxy for mineral density, was higher after the glass (230.42) than after the stannous paste (222.06) or in untreated dentine (196.37) — a study whose authors include one at the manufacturer of the glass paste tested⁶.

The most useful thing anybody has written about this is an in vitro methodological paper from 2005, which set out to build a reproducible digital measurement on dentine discs and observed on the way that most electron-microscope work claiming to quantify tubule occlusion was in fact descriptive, qualitative or semi-quantitative¹¹. It also contains the detail that should make anyone cautious about a single figure. Applying two defensible statistical treatments to the same in vitro micrographs, the authors found that multilevel modelling showed tubule area and diameter falling after one application (P < 0.001) while single-level analysis of the same images showed them rising¹¹. The pictures did not change. The arithmetic did.

How were these studies actually done?

Substrate decides what a number means, and this literature runs on five substrates that behave differently.

Human dentine discs from extracted teeth are the workhorse: premolars sliced to about a millimetre, etched to open the tubules, then brushed by hand or by machine¹. Bovine dentine is used when a study needs many near-identical specimens, as in the in vitro arm that reported 96.0% of tubules occluded at three days and 99.7% at seven, in work funded by a toothpaste manufacturer⁴. Ovine dentine appears in unpublished industry work, including S3's own, which is discussed further down. Silicone replica impressions are the only way anyone has read occlusion off a tooth that is still in a head: a mould is taken of the sensitive tooth, and the mould, not the tooth, goes under the microscope in a randomised design¹⁹. In situ appliances sit between the two, holding real dentine samples in a real mouth while the wearer eats and drinks; the in situ crossover that used them ran 28 subjects over four days, with a co-author affiliated to the manufacturer of the paste that won it¹⁴.

Then there is the protocol, and no two of these are alike. One in vitro comparison treated its discs for two minutes a day for seven days¹. A second brushed twice daily for fourteen³. The nano-hydroxyapatite rinse that came first on surface coverage in vitro got a single thirty-second application and no durability test at all⁵. Some studies store specimens in artificial saliva between treatments and pour acid on the result afterwards, as in the in vitro model where a fluoride paste's deposit was washed clean out of the tubules¹⁶. Most apply no brushing force at all.

None of it has saliva in the way a mouth has saliva, a biofilm, a tongue, a diet or a person who forgets. The authors of the real-time fluid-flow experiment print the limitation themselves: their in vitro bovine dentine model ran without saliva, biofilm or mechanical brushing¹³.

Which ingredients occlude the most, on the published pictures?

The table below is ordered by the strength of the design, not by the size of the number — a randomised trial with an electron-microscope arm sits above a beaker count, however impressive the count. The last column before the funding one is the point of the exercise: what the same paper measured that qualifies its own headline.

There is no row for this brand's own reading, and that is deliberate: the work was done at the University of Reading and has never been published. The reasons are set out in full further down the page.

Read down the table and the ranking is real but narrower than it looks. Hydroxyapatite occludes dentine tubules in the laboratory, and a 2022 systematic review of 35 in vitro studies that pooled it with bioactive glass concluded that both materials do so against control⁷. It comes first in the only in vitro study that put the three main actives against each other with a stated percentage¹, and first of four marketed pastes in a second in vitro count, at 66.13%, in a study whose test paste was donated by the maker of the hydroxyapatite it contained². But in the first of those the gap between nano-hydroxyapatite and calcium sodium phosphosilicate did not reach significance in vitro (p = 0.235)¹, and in the second the leading two pastes did not separate either². The systematic review that pooled the whole field declined to rank the two materials at all⁷, and recorded that it found no clinical occlusion studies to include⁷.

We repeated the search that would have overturned this, because a page that ranks ingredients should say what it looked for. A PubMed search for a head-to-head electron-microscope comparison containing hydroxyapatite together with stannous fluoride, a bioactive glass or arginine, restricted to records reporting a percentage or a quantitative image analysis, returned three results on the tenth of September 2026, one of which was the in vitro study already at the top of this section¹. Widening it to drop the quantification clause returned ten records, and none of them ranks three actives with stated percentages. So the honest form of the headline is this: one published in vitro study has ranked the main occluding ingredients side by side with a number attached, and its top two were not significantly different from each other¹.

There is a second thing this page does not use. Some electron-microscope comparisons of desensitising toothpastes appear in journals with no independent indexing at all — not in MEDLINE, not in PubMed or PMC, not in DOAJ, not in Scopus — and none of them is cited here, however useful the picture. Stating the rule seems better than applying it quietly.

Potassium nitrate is the ingredient this literature cannot agree about. One in vitro electron-microscope study of 90 dentine samples found a potassium nitrate toothpaste producing detectable occlusion at three, seven and fourteen days, more from the paste than from a mouthwash⁸. A second in vitro study, running a four-day acid model on 45 disks, found both dipotassium oxalate and potassium nitrate occluding more than an untreated control¹⁰. Against them sits a double-blind randomised trial of 30 adults whose companion electron-microscope arm showed occlusion with the glass and none at all with 5% potassium nitrate⁹. And in the in vitro rig described above, a potassium nitrate rinse was used as the negative control for occlusion, on the assumption that it does not occlude⁵. Two laboratory experiments point one way; one randomised trial with a microscope attached, and one rig that treated the answer as settled before it started, point the other. This page reports that as a contradiction and does not average it away; our page on whether potassium nitrate blocks dentine tubules works through the disagreement in detail.

What do the same papers measure that the ranking does not survive?

This is the section the page exists for. Every paper in the table above printed a second measurement, and the second measurement is where the ranking comes apart.

Fluid flow. The hydrodynamic account of sensitivity is about movement of fluid in the tubule, so the closest laboratory proxy for the mechanism is not a photograph but a flow rate. The in vitro study that ranked a nano-hydroxyapatite rinse best on surface coverage put the same 25 discs on a fluid-flow rig, where the oxalate and arginine/PVM-MA copolymer rinses reduced flow more than the nano-hydroxyapatite rinse did, the copolymer most of all⁵. Two instruments, one experiment, two different winners. A 2026 in vitro study measured flow in real time through 60 bovine dentine specimens under simulated pulpal pressure and found four desensitising toothpastes cutting it by 44.07% at the weakest and 69.99% at the strongest, with the other two between them, against 60.01% for a professionally applied sodium fluoride varnish¹³. Two caveats travel with that in vitro figure, and the authors print both. One of the four pastes did not beat distilled water¹³. Distilled water on its own cut measured flow through the bovine specimens by about 30% while it sat on the surface, and flow rose again as soon as it was removed¹³. There the microscope and the flow rig agreed with each other. In the older study they did not, and nobody has explained why.

Hardness and elemental analysis. The fourteen-day in vitro comparison of three desensitising pastes ran two more measurements alongside its occlusion score, and neither flattered the winner. Dentine hardness did not differ between any of the groups in vitro (p = 0.372)³. Energy-dispersive X-ray analysis found calcium and phosphorus significantly raised in two of the three treated groups in the same in vitro test — and not in the zinc hydroxyapatite group, which had the best occlusion score of the three³. A picture of a filled tubule and a measurement of what filled it are not the same evidence, and here they pointed different ways.

Acid. A tubule plugged in a beaker meets orange juice in a mouth. Two designs have tested that directly. In an in situ crossover, 28 people wore appliances carrying dentine samples brushed with a strontium acetate paste or an arginine paste and drank agitated grapefruit juice on two of the four days; both actives occluded significantly better than water and the control paste on day two, and after the challenges the strontium paste occluded better than everything else, in a study with a co-author affiliated to the manufacturer of the paste that won¹⁴. The authors of that in situ study describe the arginine paste as the more susceptible of the two to acid, and it is a manufacturer-affiliated paper saying so about a rival's active¹⁴. A confocal comparison in the laboratory watched the same active lose its lead. Before any acid was applied, the arginine paste showed 72.25% of tubules sealed in vitro and the calcium sodium phosphosilicate paste only 49.9% (p < 0.001)¹⁵. After a citric-acid challenge the arginine figure fell in vitro to 42.55% while the glass barely moved at 43.15%, and the two ended level (p = 0.901)¹⁵. Which of the two you would call the better occluder depends entirely on whether the reading was taken before lunch or after it. A third in vitro study reported that a fluoride paste's particles were washed clean out of the tubules by acid while the hydroxyapatite and bioactive glass deposits stayed¹⁶. In that same in vitro paper, the fluoride paste produced the largest rise in pH of the group¹⁶.

Time. Nobody has measured how long a toothpaste's plug lasts in a human mouth. The nearest published measurement is in dogs: an in vivo study of a calcium phosphate precipitation method found most tubules still occluded with an apatitic precipitate seven days after treatment where the teeth were brushed daily, and found them reopening with no precipitate left where no plaque control was carried out¹⁷. The material is a professionally applied treatment on cut, etched dentine, and dogs are not people. What it establishes is narrower and more useful than a durability claim: what happens to the plug depends on what happens in the mouth afterwards.

And the ruler itself. Everything above assumes the numbers are commensurable, and they are not. A percentage counted at two thousand times magnification in one laboratory, a five-grade score judged by three trained examiners in another, a depth in microns from a block-face microscope in a third — these measure different things, and the in vitro methodological study that examined the field most closely concluded that most quantitative-sounding electron-microscope work on occlusion was descriptive, qualitative or semi-quantitative¹¹. There is no validated occlusion index that laboratories share. When you meet two occlusion percentages from two papers, the safe assumption is that they cannot be compared. The clinical instruments have the same problem in a different key, and our page on how desensitising toothpastes are tested works through the Schiff score, the tactile probe and the visual analogue scale.

Does a sealed tubule mean a comfortable tooth?

The honest answer is that nobody has shown it, and the two careful attempts to show it disagreed with each other.

One research group in Bristol has taken silicone replicas of the same sensitive teeth before and after four weeks of use, twice. The first attempt was an examiner-blind randomised trial of 19 participants with a co-author employed by a toothpaste manufacturer, and in it the occlusion scores did not correlate significantly with the pain scores¹⁸. In that same manufacturer-linked randomised trial the fall in occlusion score across the four weeks did not reach significance with either paste (p = 0.0625), and the thermal rating dropped significantly from baseline for the paste that was not supposed to occlude anything¹⁸. The group's second randomised trial, 20 participants, reported the correlation as significant, with a significant fall in pain only in the occluding arm¹⁹. Its authors call the replica technique a proof of concept needing refinement before it can quantify occlusion in the mouth reliably¹⁹. And the manufacturer-linked randomised trial raises a possibility that ought to unsettle anyone reading a replica image at all: the impression material may itself have sheared off into the tubules and produced false readings of occlusion¹⁸. Two results, one group, one technique, opposite answers. Averaged, that pair would say nothing true. Our page on the hydrodynamic theory of tooth sensitivity sets out what the mechanism predicts and where the predictions have and have not been tested, and the page on tubule occlusion versus nerve desensitisation compares the two approaches as approaches.

Three further findings point the same way. A randomised trial of 45 patients applied hydroxyapatite in the surgery and saw tubules completely obliterated in the laboratory arm, with significantly less hypersensitivity than water or no treatment at one day and at one, two and four weeks (P < 0.001)²⁰. In that same randomised trial, at eight weeks the four groups no longer differed from one another²⁰. The one paper that pairs a very high laboratory count with a clinical arm — 99.7% of bovine tubules at seven days — reported in that arm a 21.39% improvement in the air-blast score at six weeks and 29.86% at eight against placebo, in a double-blind randomised trial funded by a toothpaste manufacturer with four of its six authors employed by that company⁴. The same manufacturer-funded randomised trial describes its clinical result as comparable to its bioactive-glass positive control rather than better than it⁴. Both of those figures are true, they sit in the same manufacturer-funded paper, and only one of them is about a person⁴.

Set against the clinical literature, the microscope ranking and the trial ranking are not the same ranking. A 2026 systematic review and network meta-analysis of 93 randomised trials in 9,548 participants put stannous fluoride and arginine forward as first-line self-care options at two weeks and rated the stannous cold-air result high confidence²¹. In that same network meta-analysis nano-hydroxyapatite carried the larger point estimate on the two-week cold-air score (MD −0.96, 95% CI −1.40 to −0.52) at moderate confidence, resting on two studies²¹. A 2020 network meta-analysis of 125 randomised trials in 12,541 patients put calcium sodium phosphosilicate top across all three stimuli, with potassium and hydroxyapatite together at SMD 2.47 on tactile and 2.44 on air against a fluoride toothpaste²². A 2023 systematic review and meta-analysis of 44 clinical trials, whose authors include scientists employed by a maker of hydroxyapatite toothpastes, put hydroxyapatite products 39.5% ahead of placebo and 23% ahead of fluoride toothpaste²³. That same manufacturer-affiliated meta-analysis found the difference against other desensitising agents not statistically significant (10.2%, 95% CI −19.26 to 21.76)²³. And a 2019 systematic review and meta-analysis of six randomised trials found nano-hydroxyapatite ahead of comparators on evaporative and tactile stimuli and level with them on cold (SMD −0.17, p = 0.61)²⁴.

Put the two rankings beside each other. Under the microscope, hydroxyapatite leads and the glass is close behind. In the randomised evidence, the glass leads one network meta-analysis²² and stannous fluoride leads another²¹, hydroxyapatite's two-week estimate rests on two studies²¹, and the manufacturer-affiliated meta-analysis of hydroxyapatite could not separate it from the other desensitisers²³. A brand that sells hydroxyapatite has an obvious interest in the first ranking. The second is the one that measures people.

What the microscope cannot see.

This page compares ingredients and stated actions only, not clinical performance, based on what each brand states about its own formula. Prices and formulations may change; always check the pack.

Where does S3's own laboratory reading sit in this evidence?

Outside it, and on purpose.

The formula has been tested at the University of Reading, where an electron-microscope and elemental reading compared it with a named competitor. That reading is not published. It was made on ovine dentine, it used one tooth per formulation for the quantification, and what went under the microscope was a prototype, not the tube on sale. Every one of those conditions would have to be printed beside any figure taken from it, and a page whose whole promise is a weighed reading of the published evidence does not get to slip its own unpublished favourable comparison into its own ranking table. Readers who want that work discussed will find it on the page about how desensitising toothpastes are tested, which has a section on how this one has and has not been tested.

What belongs here instead is what a reader can check against the tube and against the papers above. Two hydroxyapatites sit in the tube at two particle sizes, and the sizes are the argument: rod-shaped, crystalline and under 100 nm for the inside of the channel, around two microns and less crystalline for the surface it opens onto. The figures on the pack are inclusion levels of the ingredient as supplied and the active hydroxyapatite content is lower, which the brand states in both forms. The division of labour is the whole point of the formula: potassium at the nerve, the nano form inside the tubule, the biomimetic form across the surface, with none of the three able to cover for the others. How the filling happens is the subject of our page on how nano-hydroxyapatite seals a dentine tubule, and why a formula carries two forms of hydroxyapatite at all is argued on its own page. The fluoride is 1450 ppm as sodium monofluorophosphate, chosen because it does not react with the calcium in hydroxyapatite inside the tube — a genuine formulation constraint, and one reason hydroxyapatite pastes so often arrive with no fluoride at all. The ingredients are supported by more than 90 verified published studies, and this page has spent several thousand words on what that kind of evidence can and cannot settle. The consumer figures that appear elsewhere on the brand's site come from a panel of 51 adults reporting how their teeth felt, which is a different instrument again, and not a tubule measurement of any kind.

On the UK shelf the same division runs through the products themselves. Sensodyne Repair & Protect carries calcium sodium phosphosilicate; Sensodyne Rapid Relief carries stannous fluoride; Biorepair Plus Sensitive carries zinc hydroxyapatite; Bioniq Repair carries hydroxyapatite with no fluoride stated on the pages read; Colgate Sensitive Repair & Prevent publishes arginine at 8.00%, one of the very few rows in the whole scan to print its active's level. S3 is the row carrying potassium nitrate, both hydroxyapatites and monofluorophosphate together. If cold drinks are the trigger that brought you here, what happens inside a dentine tubule when you drink something cold is the page to read next, and the Journal's how sensitive toothpastes work, and why yours might not be working for you is the short version of this whole argument.

Frequently asked questions

Does a higher occlusion percentage mean a better toothpaste?

No, and the honest reason is that the percentages are not comparable with one another. They come from different substrates, different treatment times, different magnifications and different scoring methods, and the in vitro methodological work that examined those methods most closely found most of the field to be descriptive, qualitative or semi-quantitative rather than genuinely quantitative¹¹. Within one paper a comparison is meaningful. Across papers it is guesswork dressed as arithmetic.

Is nano-hydroxyapatite better than NovaMin at sealing tubules?

Not on the published evidence. In the one in vitro study that compared the three main actives directly, nano-hydroxyapatite occluded 98.1% of tubules and calcium sodium phosphosilicate 83.1%, and the gap between those two did not reach statistical significance (p = 0.235), while the gap to arginine did¹. A 2022 systematic review of 35 in vitro studies pooled both materials and reported that each occludes tubules against control, without ranking one above the other⁷. Hydroxyapatite is the ingredient this brand sells, and the answer is still no.

Do these laboratory results happen in a real mouth?

Partly, and much less tidily. Occlusion has been read off living teeth using silicone replica impressions in two small randomised trials¹⁹. It has also been read from dentine samples carried through real meals on an intra-oral appliance, in an in situ crossover with a co-author affiliated to a toothpaste manufacturer¹⁴. What those designs add is discouraging for anyone quoting a beaker figure. In the manufacturer-affiliated in situ crossover, two acid drinks undid one active's plug and left the other's largely intact¹⁴. And the two replica-based randomised trials that set occlusion against pain reached opposite conclusions, one of them manufacturer-affiliated¹⁸¹⁹.

Has S3 been shown to seal tubules in people?

No. The Reading work is a laboratory reading on ovine dentine, with one tooth per formulation and a prototype formula, and a laboratory reading is not a measurement in a mouth. The clinical evidence on this page is evidence for ingredient classes — hydroxyapatite, potassium, the glass, stannous fluoride, arginine — and not for any finished tube, this one included. The figures quoted elsewhere on the brand's site come from a panel of 51 adults reporting how their teeth felt over eight weeks, which measures perception rather than dentine.

How long does a sealed tubule stay sealed?

Nobody has published the answer for a toothpaste in a human mouth. The longest in situ measurement in this set ran four days, in a manufacturer-affiliated crossover¹⁴. The only duration study is an in vivo one in dogs, where most tubules were still occluded at seven days if the teeth were brushed daily and reopened if they were not¹⁷. In the randomised trial of surgery-applied hydroxyapatite, the treated group's advantage over the controls had gone by week eight²⁰. All of which points back to the plain version from the Oral Health Foundation: the effect is maintained only while you keep using the toothpaste²⁵.