Named for the Moon

Two hundred and one people were poisoned in 2008 by a liquid supplement that contained two hundred times the selenium on its label. That is the easy story. The harder one is that selenium is genuinely essential — your body reassigns a stop codon to build it into twenty-five proteins, and without it a benign virus can mutate into a heart-destroying one. This episode is about an element named after the moon, which makes no light of its own, and about the trial that launched an industry by answering a question it was never designed to ask.

Show notes

In the spring of 2008, a woman in Tennessee noticed her hair was coming out. Not thinning. Coming out — in the shower, on the pillow, in her hands. Her fingernails had gone strange first, ridged and discoloured, and then they began to lift away from the nail bed. She was tired in a way that sleep did not touch. She had diarrhoea she could not explain. And there was a smell she could not place until somebody else named it for her, because you cannot smell it on yourself: garlic. On her breath, in her sweat, constantly.

She had started taking a liquid dietary supplement a few weeks earlier.

By the time the Centers for Disease Control finished counting, there were two hundred and one people across ten states with the same story. One was hospitalised. The investigators defined a case as anyone who developed symptoms of selenium toxicity within two weeks of taking a product made by a single manufacturer and bought after the first of January that year.1

The product was a liquid supplement. It had a label. The label was wrong. It contained two hundred times the concentration of selenium it claimed.

The median estimated daily dose those two hundred and one people were swallowing was 41,749 micrograms. The recommended dietary allowance for an adult is fifty-five.

Here is what that did to them, in the investigators' own tally. Diarrhoea in 78%. Fatigue in 75%. Hair loss in 72%. Joint pain in 70%. Fingernails discoloured or brittle in 61%. Nausea in 58%.1

And here is the part I want you to hold on to, because it is the part that gets left out when this story is told as a cautionary anecdote. Ninety days later, it had not resolved. More than half still had fingernail discoloration or nail loss. A third were still exhausted. Nearly a third were still losing hair. Three months after they stopped.

Eight of them had serum selenium measured. The mean was 751 micrograms per litre. The laboratory's reference ceiling is 125.1

The authors of that paper end it with one sentence that I think about a lot. They write that had the manufacturers “been held to standards used in the pharmaceutical industry, it may have been prevented.”1

That is a careful, federal, peer-reviewed way of saying: this is a regulatory gap and everyone knows it.

But it is not the story I want to tell you today. That story — supplement manufacturing is unregulated, things go wrong — is true, and it is also easy, and easy stories make people feel informed without making them any better at deciding what to do on a Tuesday morning in front of a shelf.

The story I want to tell you is harder, and it is this: selenium is genuinely, unambiguously essential. You will die without it. The biology it sits inside is one of the most remarkable things I know about how a cell reads its own instructions. The deficiency diseases are real, they are severe, and they have killed people within living memory.

And almost none of that is an argument for the bottle.

This is the episode about the gap between your body requires this and you should buy this, and selenium is the element where that gap is narrowest and most dangerous. So let us start with the name, because the name already tells you the shape of the problem.

An element named after the moon

In 1817 the Swedish chemist Jöns Jacob Berzelius was investigating a red sludge that kept accumulating in the lead chambers of a sulfuric acid works at Gripsholm. He thought it was tellurium, an element named a few decades earlier from the Latin tellus — the earth, the ground, the soil.

It was not tellurium. It was something new, and it sat directly above tellurium in what would become the periodic table — the same family, the same chemistry, a lighter sibling. So Berzelius named it for the companion of the earth. Selene. The moon.

I want to sit with that for a second, because it is not decoration.

The moon makes no light. Not one photon of what you see on a clear night originates there. The moon is a reflector — it takes something else's output and returns it, and it is magnificent, and it is entirely dependent.

Selenium, in your body, does nothing on its own either. As a free element it has no biological activity worth the name. It is not an antioxidant in the sense the bottle means — it does not go around your bloodstream neutralising things. It has exactly one job: to be built into a protein, and to sit at that protein's active site doing chemistry that sulfur, its nearest chemical relative, cannot quite manage.

Take the protein away and selenium is a metalloid with an unpleasant smell. That distinction — between the element and the machine it is a part of — is going to do a great deal of work later on.

The stop sign that means keep going

Now the part that is genuinely beautiful, and I am going to take it slowly because it is the foundation of everything after it.

Your DNA is read in three-letter words called codons. Sixty-one of them specify one of the twenty standard amino acids — the beads that get strung together into a protein. Three of them say stop. When the machinery reading the message hits a stop codon, it releases the finished protein and lets go. UGA is one of those three stop signals.

Except sometimes it is not.

When a particular hairpin-shaped fold of RNA is present further down the message — a structure with the unlovely name of a selenocysteine insertion sequence — the cell reads that same UGA not as stop but as an instruction: insert selenocysteine here and carry on.

Selenocysteine is the twenty-first amino acid. It is cysteine with the sulfur swapped for selenium, and that single substitution changes its chemistry completely: it is far more reactive, it gives up an electron far more readily, and at the pH of your cells it is already ionised and ready to work where cysteine is not.

So the cell has taken a full stop and, conditionally, turned it into a comma. Same three letters. Opposite meaning. Everything depends on what is downstream.

It is an expensive trick. The cell has to maintain a dedicated transfer RNA, a dedicated elongation factor, and a dedicated binding protein, all for this one amino acid. Evolution does not pay that bill for a rounding error.

In 2003, a group at the University of Nebraska and the Centre for Genomic Regulation in Barcelona did the computational work to find all of them, because — and this is a lovely detail — standard gene annotation software sees UGA, calls it a stop, and truncates the gene. Most selenoprotein genes were misannotated in the reference genome. They had to be found by looking for the RNA hairpin instead.

The answer: the human selenoproteome consists of twenty-five selenoproteins.2

Three families of them matter for everything that follows.

The glutathione peroxidases. Eight of them in humans. Hydrogen peroxide is a normal by-product of using oxygen — your mitochondria make it all day — and left alone it will react with iron and produce the hydroxyl radical, which is the genuinely destructive one, the one that breaks DNA. Glutathione peroxidase takes peroxide and turns it into water. Selenocysteine is the atom that does it.

The thioredoxin reductases. Three of them. These keep a separate family of proteins in their reduced, working state — think of them as the service crew for a large part of your cell's redox machinery, including the enzyme that makes the building blocks of DNA.

And the deiodinases. Three of them, and this is the one people have heard of without knowing why.

Your thyroid mostly makes T4 — thyroxine, four iodine atoms. T4 is not very active. It is a circulating reservoir. The hormone that actually goes into the nucleus of your cells and changes which genes are switched on is T3, with three iodines. The conversion from one to the other is the removal of a single iodine atom, and the enzymes that remove it — type 1 and type 2 deiodinase — are selenoproteins. Type 3 deiodinase, also a selenoprotein, does the opposite: it inactivates thyroid hormone when you have had enough.3

So when a bottle says “supports thyroid function”, that is not a lie. It is a real mechanism. Whether it is a reason is a completely different question, and we will get there.

What happens when there is none: Keshan

In 1935, in Keshan County in Heilongjiang — the far north-east of China, near the Russian border — physicians recorded an outbreak of a heart disease nobody could explain. It struck children and women of childbearing age hardest. The heart would dilate and fail. In the acute form people could die within days. It came in waves, it was fiercely seasonal, and it followed a geographic belt that ran diagonally across China from the north-east to the south-west.

They named it after the county. Keshan disease.

The belt turned out to map onto something in the ground. The soil along it is selenium-poor, and in a subsistence agricultural economy the selenium in your body is the selenium in the soil that grew your grain.

In 1979 the Chinese Medical Journal published the intervention. Sodium selenite — a simple inorganic selenium salt — given to children in the affected areas.4 The disease went away.

That is a clean result and I want to be precise about what it establishes and what it does not. It establishes that severe selenium deficiency causes a fatal cardiomyopathy and that replacing the selenium prevents it. It is one of the clearest nutritional causation stories in twentieth-century medicine.

It establishes nothing whatsoever about what happens if you already have enough. Restoring a person from nothing to something is a different experiment from moving a person from enough to more. That sentence is the whole episode, and I am going to say it again later.

But there is a second half to Keshan that almost nobody tells, and it is the strangest thing in this entire script.

Keshan disease was seasonal. Selenium in soil is not seasonal. Something else was coming and going, and the suspicion for decades was a virus — specifically coxsackievirus B, an enterovirus that causes myocarditis, inflammation of the heart muscle.

So which was it? The deficiency or the virus? For a long time that was framed as a competition between two hypotheses.

In 1995 a group led by Melinda Beck showed it was never a competition.

Here is the experiment. Take a strain of coxsackievirus B3 that has been cloned and sequenced and is known to be amyocarditic — benign, it does not damage hearts. Infect two sets of mice: one selenium-adequate, one selenium-deficient.

The deficient mice got myocarditis. That much you might predict — a weakened host, a worse infection.

Then they did the step that changes everything. They recovered the virus from the hearts of the deficient mice and injected it into selenium-adequate mice. Normal mice. Well-nourished mice.

Those mice got myocarditis too.5

The virus had changed. Sequencing found six nucleotide changes between the benign virus that went in and the virulent virus that came out, and the changes were consistent with the known base composition differences between virulent and avirulent coxsackievirus strains. It was not a sicker host. It was a different virus, and the host's nutritional state had made it.5

The authors' own summary: to the best of their knowledge, this was the first report of a specific nutritional deficiency driving changes in a viral genome, letting an avirulent virus acquire virulence through mutation.5

Later work pinned the mechanism down further: mice engineered to lack glutathione peroxidase-1 — one selenoprotein, knocked out — develop the same myocarditis from the same benign virus. It is not selenium in the abstract. It is that specific enzyme, doing that specific job.6

I find this genuinely unsettling in the best way. We are used to thinking of nutrition as something that happens to us — our immunity, our resilience, our ability to fight something off. This says the deficiency reaches past you and edits the pathogen. A poorly-nourished population is not just more likely to get sick; it is a more permissive environment for viral evolution.

Keep that in your pocket. It is the strongest case anybody will ever make for selenium, and I want you to have the strongest version of the case before I start taking it apart. Because here is the thing about the strongest version: it is a case for not being deficient. It is not a case for more.

What happens when there is too much: Enshi

Six hundred miles south-west of the Keshan belt, in Enshi County in Hubei Province, the opposite thing was happening at almost exactly the same time.

Starting in 1961 the villagers began losing their hair and their fingernails. In the worst-affected areas there were skin lesions, and changes in the nervous system, and possibly damage to the teeth. One middle-aged woman became hemiplegic — paralysed down one side — and died, and her illness was attributed to the poisoning.

Across the five most heavily affected villages, in a population of two hundred and forty-eight people, morbidity approached fifty per cent during the peak years of 1961 to 1964.7

When the dietary intakes were finally estimated, after the worst had passed, they averaged 4.99 milligrams of selenium a day, with a range from 3.2 to 6.7.7

Let me put that beside the number from earlier. The recommended dietary allowance is fifty-five micrograms. Five milligrams is five thousand micrograms. These villagers were eating roughly ninety times the recommended intake, every day, for years.

Their hair selenium averaged 32.2 micrograms per millilitre and their blood 3.2. Comparing vegetables, cereals, hair, blood and urine from the selenosis areas against the Keshan deficiency areas, the investigators found differences of up to a thousandfold — in the same country, in the same decade.7

Where was it coming from? Coal.

The bedrock under Enshi includes a stony coal with an average selenium content above 300 micrograms per gram. One sample they measured exceeded eighty thousand. The coal weathered, the selenium entered the soil, and — this is the agricultural detail that completes the chain — the traditional local practice of liming the fields made that selenium chemically available for crops to take up.7

And the trigger for the outbreak itself was a drought. The rice crop failed. Rice is relatively low in selenium. With no rice, the villagers ate more high-selenium vegetables and maize, and fewer protein foods.7

Geology to agriculture to crop failure to hair falling out in a farmhouse. That is the whole causal chain, and every link in it is ordinary.

The word for this is selenosis, and the clinical picture is worth memorising because it is the same picture the two hundred and one Americans had in 2008, just arrived at from a bottle instead of from a coal seam: hair loss, nail loss and discoloration, gastrointestinal upset, fatigue, a garlic odour on the breath, and — with enough exposure for long enough — peripheral neuropathy and neurological change.

Same element. Same molecule, mostly: the selenium in that Enshi maize and rice was analysed and found to be predominantly selenomethionine — which is also the form in the supplement aisle, and also the form in a Brazil nut.

Why the curve is a U and not a slope

So we have two real diseases at two extremes, and now the obvious question. Where is the line? And why is there a line at all — why does a nutrient stop helping and start hurting, rather than just stopping helping?

This is where selenium becomes genuinely instructive, because for once we know the mechanism, and the mechanism is almost embarrassingly simple.

Meet selenoprotein P. It is made in your liver, it is the only selenoprotein that carries multiple selenocysteines — about ten of them — and its job is not enzymatic at all. It is a delivery van. It carries selenium out of the liver and distributes it to the tissues that need it, with the brain and the testes at the front of the queue.

In 2010 a group at the University of East Anglia ran the experiment that tells you where the line is. A hundred and nineteen healthy UK adults aged fifty to sixty-four, randomised, double-blind, placebo-controlled, twelve weeks, on selenium-enriched yeast at 50, 100 or 200 micrograms a day.8 They measured plasma selenium, platelet glutathione peroxidase activity, and selenoprotein P.

Plasma selenium did exactly what you would expect: it went up with the dose, and kept going up. Baseline 95.7 nanograms per millilitre, rising to 118 on 50 micrograms, 152 on 100, and 177 on 200.8

Selenoprotein P did something else entirely. It rose from a baseline of about 5.0 micrograms per millilitre to 6.17 on fifty micrograms a day — and then, on a hundred, 6.73. And on two hundred, 6.59.8

Read those last two numbers again. Doubling the dose from 100 to 200 did not raise selenoprotein P. If anything it drifted down within the noise.

And platelet glutathione peroxidase activity did not change significantly at any dose or in any form.8

That is a plateau, and it is the single most important fact in this episode.

The authors' conclusion was that in this population, about fifty micrograms a day on top of a habitual intake of roughly fifty-five was what it took to optimise selenoprotein P.8 In other words: somewhere near a hundred micrograms a day total, the system is full. The vans are all loaded. The enzymes are all built.

So what happens to the next microgram?

It gets filed in the wrong place, and this is the mechanism that makes the curve a U.

Selenocysteine — the functional one, the one at the business end of your twenty-five selenoproteins — can only be made by that dedicated, regulated, expensive machinery I described earlier. The cell decides how much to make.

Selenomethionine cannot be regulated that way. It is not recognised as special. It is structurally close enough to ordinary methionine that your protein-building machinery, when it reaches a codon calling for methionine, will grab whichever is nearest — and if your selenomethionine pool is large, it will sometimes grab that one instead.

So it goes into your haemoglobin, your muscle, your albumin, your keratin. Not as an enzyme. Not doing chemistry. Just there, substituted at random, in proportion to how much you ate.

That pool is the toxic pool. It has no ceiling and no thermostat. It builds up in the tissues with the fastest protein turnover, which is exactly why the first things to go are hair and nails — keratin is enormously rich in sulfur-containing amino acids, and selenium is sulfur's understudy.

Once you understand that, the whole shape falls out. Below saturation, selenium is limiting and more of it buys you more functioning enzyme. Above saturation, more of it buys you nothing at all — and starts depositing an unregulated substitute into structural proteins that never asked for it.

The benefit has a ceiling. The exposure does not. That is a U-shaped curve, and it is not a statistical artefact. It is biochemistry.

And you can see it in a population. In the third National Health and Nutrition Examination Survey, serum selenium was measured in 13,887 American adults and they were followed for up to twelve years.

Mean serum selenium: 125.6 nanograms per millilitre. Mortality fell as selenium rose — up to about 130. Above about 150, it rose again.9

The authors put it carefully: a nonlinear association, with an inverse relationship at low levels and “a modest increase in mortality at high selenium levels”.9

Look at where the American average sits on that curve. One hundred and twenty-five. The population mean is already at the bottom of the U, on the flat part, about to start climbing.

The trial that built an industry by missing

Now the cancer story, which is the reason most people have ever heard of selenium, and which is a masterclass in how an honest trial becomes a dishonest marketing claim without anybody lying.

In 1983, Larry Clark and colleagues began enrolling patients across seven dermatology clinics in the eastern United States. The hypothesis was that selenium would prevent skin cancer. They recruited 1,312 people with a history of basal cell or squamous cell carcinoma, randomised them to 200 micrograms of selenium daily as selenised yeast or placebo, and followed them.

The trial was called the Nutritional Prevention of Cancer trial — NPC. It ran for a mean of four and a half years of treatment and 6.4 years of follow-up.10

It failed.

I want to be very plain about this because it is routinely glossed. The primary endpoints — the outcomes the trial was designed around, powered for, and registered to test — were the incidence of basal cell and squamous cell skin cancer. The results were a relative risk of 1.10 for basal cell (95% CI 0.95–1.28) and 1.14 for squamous cell (0.93–1.39).10

Both above one. Neither statistically significant. Selenium did not prevent the cancer it was given to prevent, and the point estimates, if anything, lean the wrong way.

But in 1990 — seven years after enrolment began — the investigators added a set of secondary endpoints: all-cause mortality, total cancer mortality, total cancer incidence, and the incidence of lung, prostate and colorectal cancer.10

And those came back spectacular.

Total cancer mortality: 29 deaths on selenium against 57 on placebo. Relative risk 0.50. Total cancer incidence: 77 against 119. Relative risk 0.63. Significant reductions in lung, colorectal and prostate cancer individually.10

The blinded phase was stopped early, primarily because of those findings.10

You can feel the gravity of it. A cheap, natural, non-toxic mineral, halving cancer deaths in a randomised controlled trial. That is a career-defining result and it was published in JAMA in 1996.

Here is the sentence the authors ended on, and I am going to read it verbatim, because it is the most responsible sentence in the paper and it is the one that vanished:

“These effects of selenium require confirmation in an independent trial of appropriate design before new public health recommendations regarding selenium supplementation can be made.”10

They knew. They knew exactly what they had and exactly what they did not have, and they said so in the abstract.

So why is that caution justified? Why is a halving of cancer mortality not enough on its own?

Because of what a secondary endpoint is. When you design a trial you pick one outcome and you power the study to detect a change in it, and you accept a one-in-twenty chance of being fooled by randomness. If you then measure twenty other things, you should expect one of them to look significant by chance alone. The protection that randomisation gives you applies to the question you asked. It does not stretch to cover every question you thought of afterwards.

And these were added seven years in, in a trial that had already failed its actual endpoint, in a population selected for a history of skin cancer, in a region of the eastern United States chosen specifically because selenium intake there was low.10

None of that makes the result wrong. It makes it a hypothesis — a very good, very interesting, very fundable hypothesis. Clark said so himself.

The supplement industry did not read it as a hypothesis. It read it as a finding, and the 1996 NPC secondary endpoints are, functionally, the reason there is a selenium section in your pharmacy.

Before we leave NPC there is a follow-up analysis that matters enormously for where this ends up. In 2003 the group published the complete blinded treatment period, looking specifically at prostate cancer and at where people started.

Overall prostate cancer relative risk: 0.51. Still halved.11

But the protective effect was confined to participants in the lowest two tertiles of baseline plasma selenium — below 123.2 nanograms per millilitre. In the top third, nothing. The interaction between baseline selenium and treatment was statistically significant.11

Hold that number. 123.2. Remember the American population mean: 125.6.

The confirmation trial nobody wanted

Clark asked for an independent trial of appropriate design. He got one, and it was enormous.

SELECT — the Selenium and Vitamin E Cancer Prevention Trial — randomised 35,533 men at 427 sites across the United States, Canada and Puerto Rico between 2001 and 2004. Four arms: selenium alone, vitamin E alone, both, or neither. The selenium dose was 200 micrograms a day as L-selenomethionine. The planned follow-up was a minimum of seven years and a maximum of twelve.12

This is about as good as cancer prevention research gets. Enormous, blinded, placebo-controlled, four-armed, pre-registered, publicly funded.

It was stopped early.

At a median follow-up of 5.46 years, the hazard ratio for prostate cancer was 1.04 on selenium (99% CI 0.87–1.24). On vitamin E, 1.13. On both, 1.05. No significant differences in any other prespecified cancer endpoint. The conclusion in the abstract is one sentence long: selenium or vitamin E, alone or together, at these doses and formulations, did not prevent prostate cancer in this population.12

The 1996 signal did not replicate. Not attenuated. Not “positive but smaller than hoped”. Absent.

And then the follow-up got worse.

In 2011 the investigators published a further 54,464 person-years and 521 additional prostate cancers. With the longer view, vitamin E alone significantly increased prostate cancer risk by 17% — hazard ratio 1.17, 99% CI 1.004 to 1.36.13

I want to pause on vitamin E for a moment even though this is a selenium episode, because the lesson generalises. Vitamin E was in that trial on the strength of its own promising secondary endpoint, from a lung cancer prevention trial in Finnish smokers. Same logic, same confidence, same outcome: tested properly, on its own, at scale, an antioxidant that everyone was sure was at worst harmless turned out to cause the disease it was given to prevent.

Now the finding that makes SELECT the most important selenium trial ever run, and it is not in either of those two papers. It is a 2014 case-cohort analysis in the Journal of the National Cancer Institute, and it asked a question the main trial could not: did it matter how much selenium a man already had?

They had toenail clippings from baseline. Toenail selenium is a good long-term integrator — it reflects intake over months, not the last meal. 1,739 prostate cancers, 489 of them high-grade, against a randomly selected cohort of 3,117 men.14

Three results, and they fit together into one picture.

One. In men who took no supplement, toenail selenium was not associated with prostate cancer risk at all. The observational premise — more selenium, less cancer — simply was not there.14

Two. Selenium supplementation had no effect among men in the bottom 60% of baseline selenium.14

Three. Among men above that — men who already had plenty — selenium supplementation increased the risk of high-grade prostate cancer by 91%. P equals 0.007.14

High-grade means Gleason 7 to 10. That is not the indolent kind that sits there for twenty years and never troubles anyone. That is the kind that kills men.

And the same paper found the mirror image for vitamin E: no effect in men with high selenium status, but in men with low selenium status it raised total prostate cancer by 63%, and high-grade by 111%.14

The authors' recommendation, in a cancer journal, is blunt: “Men should avoid selenium or vitamin E supplementation at doses that exceed recommended dietary intakes.”14

So here is the shape of it. The same pill, in the same trial, in men who differed only in where they started, went in opposite directions.

That is not a failed drug. A failed drug does nothing. This is a nutrient behaving exactly the way the selenoprotein P plateau said it would: useful while you are below saturation, inert at it, and harmful past it.

The U-shape stopped being an epidemiological curiosity and became a randomised finding on a hard cancer endpoint. And the men it harmed were American men with ordinary American diets — not a special population. Sufficiency is the common state here, and the trial punished people for treating it as a deficiency.

The diabetes signal

There is a second harm, and it surfaced from the same trial that started all of this, which has a certain symmetry.

In 2007 the NPC investigators published a secondary analysis of the 1,202 participants who did not have diabetes at baseline. Over an average of 7.7 years, type 2 diabetes developed in 58 people on selenium and 39 on placebo — 12.6 cases per thousand person-years against 8.4. Hazard ratio 1.55, 95% confidence interval 1.03 to 2.33.15

And once again, the effect tracked where people started. Across tertiles of baseline plasma selenium there was an exposure-response gradient, and in the highest tertile the hazard ratio was 2.70 (1.30–5.61).15

Now, I said earlier that a secondary endpoint is a hypothesis rather than a finding, and I meant it, so I have to apply that rule here too. This was a secondary outcome. The diagnoses were self-reported, though validated in most participants. The sample was mostly older and white.15

So has it held up?

Yes, and that is the difference. A 2018 systematic review and meta-analysis went looking across 50 non-experimental studies and 5 randomised trials. In the trials, selenium supplementation increased diabetes risk by 11% — RR 1.11, 95% CI 1.01 to 1.22. In the observational data there was a roughly linear dose-response: at 140 micrograms per litre of selenium exposure compared with under 45, the risk ratio was 3.6.16

I want to be careful about how I present that, because this is exactly the kind of number that gets turned into a headline it does not deserve.

An eleven per cent relative increase is small. If your ten-year risk of type 2 diabetes is five per cent, this moves it to about five and a half. On its own, for one person, that is not frightening.

The meta-analysis authors make the public health point instead, and it is the right one: the relative increase is small but of possible importance because diabetes is extremely common and selenium exposure is ubiquitous.16

The honest summary is: this is not a reason to panic. It is a reason not to take a supplement that has no demonstrated benefit for you.

What the whole literature says at once

In 2018 the Cochrane Collaboration published the third update of its review on selenium for preventing cancer. Eighty-three studies. Ten randomised trials covering 27,232 participants. Seventy observational cohort studies covering more than 2.36 million people.17

For the randomised trials at low risk of bias, the summary risk ratio for any cancer incidence was 1.01 (95% CI 0.93–1.10), at high certainty by GRADE. Cancer mortality 1.02. Colorectal 0.99. Lung 1.16. Bladder 1.07. Prostate 1.01. High certainty for nearly all of them.17

The observational studies looked better — they usually do — with an odds ratio of 0.72 for cancer incidence comparing highest to lowest selenium exposure.17 But Cochrane graded that evidence very low certainty for every outcome, and said why: exposure misclassification, and unmeasured confounding from the lifestyle and nutritional factors that travel alongside selenium status. And critically, no dose-response relationship emerged — which is one of the oldest tests we have for whether an association is causal.17

They also note, flatly, that the NPC hypothesis — that people with low blood selenium could reduce their cancer risk by taking more — “has not been confirmed”.17

The conclusion runs: “there is no evidence to suggest that increasing selenium intake through diet or supplementation prevents cancer in humans.”17

That is as settled as nutritional epidemiology gets. Twenty-two years, one hypothesis, 27,232 randomised participants, and a flat line.

Where it does work, because that matters too

I am not in the business of telling you a nutrient is useless because its biggest claim failed. That would be the mirror image of the mistake the industry made, and it would be just as lazy. So let us look at where selenium has a real, randomised, positive result — because it has one, and it is a good one.

Graves' disease is an autoimmune overactive thyroid. In some people the same autoimmune process attacks the tissue behind the eye: the eye muscles and fat swell, the eyes protrude, they ache and water and sometimes the vision is threatened. That is Graves' orbitopathy, and the mild form is genuinely miserable and has no good treatment, because the drugs that work in severe disease are steroids and radiotherapy and you do not deploy those for mild.

In 2011 the European Group on Graves' Orbitopathy published a randomised, double-blind, placebo-controlled trial in the New England Journal of Medicine. A hundred and fifty-nine patients with mild orbitopathy, given selenium 100 micrograms twice daily, or pentoxifylline, or placebo, for six months, then followed for six more after the treatment stopped.18

At six months, selenium — but not pentoxifylline — was associated with improved quality of life (P<0.001), less eye involvement (P=0.01), and slowed progression of the orbitopathy (P=0.01). Exploratory evaluation at twelve months confirmed it. No adverse events were evident with selenium.18

That is a real result and I would defend it. Note four things about it.

It was in Europe, where soil selenium and therefore population selenium status is substantially lower than in North America. It was in a specific disease with an oxidative mechanism behind it, not in healthy people hoping to prevent something. It had a defined endpoint — how the eye looked and how the patient felt. And it had a stop date: six months, and the benefit persisted after it was stopped.

Every one of those is absent from the way selenium is actually sold.

The thyroid story more broadly is messier, and I am going to show you the mess rather than pick the side I like.

Hashimoto's thyroiditis is autoimmune destruction of the thyroid, and it is extremely common, and people with it are told to take selenium constantly.

In favour: a randomised controlled trial of 90 patients published in 2021 found that six months of selenium yeast significantly reduced thyroid peroxidase antibodies, thyroglobulin antibodies and TSH, alongside rises in selenium, glutathione peroxidase-3 and selenoprotein P — and an increase in activated regulatory T cells, which is a plausible immunological mechanism.19

Against: a blinded, placebo-controlled randomised trial of 76 euthyroid Hashimoto's patients published in 2016 gave L-selenomethionine 166 micrograms daily for six months and found no difference from placebo in TSH, free T4, free T3, thyroid peroxidase antibodies, thyroid echogenicity or CXCL10. The authors' own phrasing is that their results “tip the balance toward the ineffectiveness” of short-term supplementation.20

Two randomised trials, opposite answers. That happens, and when it does you do not get to pick.

But there is a deeper problem with the Hashimoto's literature that resolving the contradiction would not fix, and it is worth naming because it applies far beyond thyroid disease.

The endpoint in almost all of these trials is antibody titre. Thyroid peroxidase antibodies are a marker. They are not a symptom. Nobody feels their TPO antibodies. What patients want to know is whether they will need levothyroxine, whether they will need more of it, whether the fatigue lifts — and those trials are much longer and much harder and mostly have not been done.

A number that moves is not the same as a person who is better. Watch for that everywhere.

Two Brazil nuts, and why food does not solve this

The obvious response to everything I have just said is: fine, skip the bottle, get it from food. That is usually the right instinct and here it is half right, so let me show you the half that is not.

The single richest food source of selenium on earth is the Brazil nut. The tree, Bertholletia excelsa, has roots that go deep and it concentrates selenium out of Amazonian soil with almost absurd efficiency.

In 2008 a group at the University of Otago in New Zealand tested it properly. New Zealand is a low-selenium country, which makes it a good place to run the experiment. Fifty-nine adults, randomised to two Brazil nuts a day, or 100 micrograms of selenium as selenomethionine, or placebo, for twelve weeks.21

The nuts won.

Plasma selenium rose 64.2% in the Brazil nut group against 61.0% on selenomethionine and 7.6% on placebo. Whole blood glutathione peroxidase activity rose more on the nuts than on either the supplement or the placebo — significantly more than the supplement, in fact.21

Two nuts. Better than a purified 100-microgram supplement, on the enzyme activity that actually matters.

So: eat the nuts, not the pill. Except.

Read the methods. The investigators intended those two nuts to provide about 100 micrograms of selenium. When they actually assayed what the participants were eating, the mean intake was 53 micrograms — with a possible range of 20 to 84.21

From the same intervention. Two nuts, measured, in a controlled trial, with a fourfold spread between the low end and the high end.

And that is the good case, because somebody weighed and assayed them. In your kitchen there is no assay. The selenium in a Brazil nut is set by the selenium in the patch of Amazon the tree is standing in, and neither the tree nor the packet has any way to tell you.

This is the point where the natural-versus-synthetic binary falls apart completely, and it is the same collapse we have watched in every episode of this show.

The selenium in the nut is selenomethionine. The selenium in the capsule is selenomethionine. Same molecule, same absorption, same non-specific incorporation into your keratin if you overdo it.

The difference between them is not purity or naturalness. It is that the capsule tells you the dose and the nut does not. For a nutrient with a fourfold natural variation and a narrow therapeutic window, the food is the less controlled delivery system. That is an uncomfortable sentence for everybody, and it is true.

Which brings us to geography, because it explains most of the confusion in this entire literature.

Your selenium status is, to a first approximation, an address. It is the soil that grew the wheat that made the bread. The high-selenium soils of the North American Great Plains produce wheat that feeds a continent; parts of China, Finland, New Zealand and the UK sit far lower.

Which means a study done in Enshi and a study done in Nebraska and a study done in Otago are not three attempts at the same question. They are three different questions that happen to use the same word.

And it means that when you read “selenium was beneficial”, the very first thing to ask is where.

Who should not take this

This show has a rule, and the rule is that if an episode argues something works, it owes you the harms in the body of the episode and not in a footnote. This episode is mostly arguing the opposite, which makes the rule more important rather than less — because “probably useless” is exactly the frame under which people take something anyway, reasoning that it cannot hurt.

It can hurt. Named specifically:

If you live in North America, you are the population SELECT harmed. Mean serum selenium in a representative US sample was 125.6 nanograms per millilitre.9 The threshold above which NPC's prostate benefit disappeared was 123.2.11 The men in whom supplementation raised high-grade prostate cancer by 91% were the upper 40% of an ordinary American cohort.14 You do not have to be unusual to be in that group. You have to be typical.

If you have type 2 diabetes, prediabetes, or a strong family history. The randomised evidence puts the increase at 11%16 and the NPC secondary analysis put it at 55%, rising to 170% in the highest baseline tertile.15 Small in absolute terms; entirely unjustified if the benefit is zero.

If you are already taking a multivitamin, a “thyroid support” blend, a “immune” formula, or eating Brazil nuts regularly. Selenium is in all of these and people stack them without ever adding up. Two Brazil nuts plus a multivitamin plus a dedicated selenium capsule is a straightforward route past 400 micrograms a day, which is the tolerable upper intake level, without a single label saying anything untrue.

If you are pregnant. Selenoprotein P delivers preferentially to the brain and the testes, selenium crosses the placenta, and the dose-response in pregnancy has not been characterised well enough for anyone to be confident about supraphysiological intakes. Deficiency matters here; excess is unstudied. That is a reason for food and a prenatal, not for an extra bottle.

How you would know it was hurting you, in plain words, because this is the part that gets left out and it is the part you can actually use.

The earliest signs are boring and get blamed on something else. A garlic smell on your breath or your skin that other people notice before you do — that is dimethyl selenide, the compound your body makes to get rid of the excess, coming out through your lungs. Hair shedding beyond what is normal for you. Nails that turn brittle, ridged, white-flecked or discoloured, or start lifting from the bed. Persistent fatigue, nausea, diarrhoea, joint pain. With longer exposure, numbness or tingling in the hands and feet — peripheral neuropathy — and irritability.17

Notice what that list looks like from the inside. Tired, achy, GI upset, hair falling out, nails bad. If you are a woman in your thirties or forties taking selenium for your thyroid, that list will be read as your thyroid disease, or as stress, or as perimenopause, by you and quite possibly by your doctor. The supplement is the last thing anyone suspects because it is the thing you are taking to feel better.

And recovery is slow. Ninety days after stopping, more than half of the 2008 cohort still had damaged nails and a third were still exhausted.1 This is not an overnight correction.

What test to ask for, before and after. Serum or plasma selenium. It is a real, ordinary, orderable test. The reference range used by the CDC laboratory in the 2008 investigation was 125 micrograms per litre or below, and the poisoned patients averaged 751.1

If you are considering selenium for a real reason, get the number first. If it comes back above about 120, the trials say the honest answer is that you have nothing to gain and a measurable amount to lose. If it comes back genuinely low, then you have a reason — and a target, which is the plateau, not a number on a bottle.

And a stop date. Nothing on this show is a thing you take forever without a reason to keep taking it. The one trial I would defend ran for six months and then stopped, and the benefit held after it stopped.18 Six months, then re-measure, then decide again. Not a subscription.

What I actually think

I came into this expecting to defend selenium more than I can.

The biochemistry is genuinely thrilling. A cell that takes a stop codon and, conditionally, reads it as continue, maintaining an entire dedicated translation apparatus for one amino acid — that is not a quirk, that is evolution paying a steep bill for something it cannot do any other way. And the Beck coxsackievirus work is, I think, one of the most under-appreciated findings in nutritional science: a deficiency in the host that reaches past the host and mutates the pathogen.5

None of that survives contact with the question should you buy this, and I am not going to pretend otherwise.

The cancer claim is not unproven. It is disproven, at high certainty, and that is a stronger statement. Twenty-seven thousand people across ten randomised trials, a summary risk ratio of 1.01 for any cancer.17 We are not waiting on more data. We got the data.

The story of how it became a claim anyway is the most useful thing in this episode, and it required nobody to lie. An honest trial missed its primary endpoint. Its investigators added secondary endpoints seven years in, found something startling, published it, stopped the trial, and wrote in the abstract that it needed independent confirmation before anyone changed recommendations.10

Everything after that was other people reading the result and skipping the sentence.

So here is the rule I would like you to take out of this, because it will serve you on the next supplement and the one after.

A trial protects only the question it was designed to ask. Everything else it noticed is a hypothesis wearing a p-value.

That is not cynicism about research. It is the actual logic of what randomisation buys you. Clark understood it perfectly. The industry that grew out of his trial did not, or did not want to.

The second thing is the U, and it is the part that generalises furthest.

We are extremely comfortable with the idea that a drug has a dose. Nobody thinks more morphine is better morphine. But we carry a completely different mental model for nutrients: essential, natural, water-soluble-so-you-pee-out- the-extra, more-is-probably-fine.

Selenium is the cleanest refutation of that model I know, because we can see the ceiling. Selenoprotein P saturates at around a hundred micrograms a day and then flatly refuses to rise further.8 That is your body saying I have enough in the only language it has. Everything past that is not nutrition. It is accumulation.

And the third thing is what the word ‘essential’ actually means.

It comes from esse — to be. An essential nutrient is one you cannot be without; your body cannot make it and so it must come from outside. That is the entire content of the word. It is a statement about a biochemical pathway you do not possess.

It is not a statement about quantity. It is not a comparative. There is no ‘more essential’. Iron is essential and will kill a child who finds the bottle. Water is essential and drowns people.

The supplement industry has spent forty years letting the word do work it was never built for, and selenium is where you can watch the damage most precisely.

So, what would I actually say to you.

If you live in the United States or Canada and you eat food: you are almost certainly replete. Do not take a selenium supplement to prevent cancer, because that has been tested in thirty-five thousand men and it does not, and in the half of them who already had plenty it made the dangerous kind more likely.14

If you have mild Graves' orbitopathy and you live somewhere with low soil selenium, the NEJM trial is real and I would discuss six months of it with your endocrinologist.18

If you have Hashimoto's, the honest answer is that the trials disagree, the endpoint they disagree about is an antibody titre rather than how you feel, and I would rather you spent the money on a thyroid panel done properly.

If you live somewhere genuinely selenium-poor, or you have had bowel surgery, or you are on long-term parenteral nutrition, or you have a malabsorptive condition — then you are in a different episode, and yes, measure it, and yes, correct it.

And if you like Brazil nuts, eat Brazil nuts. Two, not a handful. They are delicious and they work better than the pill did in the one trial that compared them.21 Just know that you are dosing from a lottery, and that the answer to a lottery is not to buy more tickets.

The element is named for the moon, and I have been thinking about why that keeps feeling right.

Not because it is beautiful and distant, although it is. Because the moon makes nothing. It has no light of its own. Everything it appears to give you it is passing on from somewhere else, and its whole effect on the earth is a matter of position — where it is relative to you determines whether you get a spring tide or a neap, a full moon or nothing at all.

Selenium is the same. It has no activity of its own. It matters entirely according to where you already are. And the pill on the shelf cannot know that, which is precisely why the shelf is the wrong place to make the decision.

Sources

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