Immunology is dizzyingly complicated; I don’t think I can even tell you how many different kinds of interleukin there are, let alone what they do. But one of my colleagues, Duygu, was researching peanut allergy, and she gave a talk where she cited a statistic that absolutely floored me.

Before 1980, she said, there were a total of zero known fatalities from peanut allergy.1

Now, it's common knowledge that the incidence of peanut allergy is on the rise, but zero?

What happened? What changed?

What we know and what we guess

The epidemiology of peanut allergy is interesting to begin with. It's not much of a problem in most of the world. Only in America—where roughly 1% of children are thought to be affected, and maybe a dozen people die annually from it—are schoolchildren forbidden from bringing a PB&J to lunch, for fear of endangering a classmate’s life.

So what makes the USA different? In a nation of immigrants, it’s unlikely to be genetics. It’s not just exposure, either…peanuts are native to the Americas,2 but they’re enjoyed around the world—albeit not often in such quantities as here. US public health guidelines nowadays recommend waiting until children are a little older, to introduce peanuts as a food, and some people have theorized that this is somehow responsible: Early life is a time when your immune system is dynamic, plastic, learning. If you don’t introduce a food early enough, the thinking goes, you start to miss your chance to teach the body “This is food, not a pathogen”.

There’s a certain appeal to that theory, and perhaps the delayed exposure isn’t helping—but those guidelines were a response to the growing peanut allergy epidemic, so there’s got to be more to the story than that.

Aside from the incidence of allergy, though, there’s one other unique aspect of Americans’ relationship with peanuts.

One glaringly obvious thing that sets us apart.

Peanut butter.

It was obvious to everyone else at the Swiss Institute, anyway.

I, the lone American in the crowd, was surprised to learn that peanut butter is not a universal phenomenon: a fish who has made it all the way to 32 years of age without ever once being told he’s wet. It might be different these days, but growing up in the ‘90s, peanut butter was like one of the four fundamental elements of childhood—and nowhere else is it considered as much of a staple food as here.

And this was the thrust of my colleague’s research: why would peanut butter be more likely than whole peanuts to lead to peanut allergy? Her hypothesis (well, really the institute director’s hypothesis, which Duygu was assigned to investigate) was that the blame lay with emulsifiers: processing aids like soy lecithin or polysorbate 80, which help keep the final product from separating into an oily top-layer and sludge on the bottom. In cell-culture based tests, a lot of emulsifiers are toxic at moderate concentrations, and there’s been speculation that they might cause “leaky gut”: disrupting the epithelial barrier of the intestine and allowing proteins—both bacterial and dietary—to slip into circulation and trigger the immune system’s invader alert.

But the data doesn't really bear this out, or at least not the data Duygu was presenting that day. Turns out exposing intestinal cells in a petri dish to the kind of emulsifiers you find in peanut butter doesn't do them much harm.

I was barely listening by the time she got around to her “Results” slide, though—I had found something more interesting.

What Happened in 1980

Regular readers will know that I’m skeptical of some modern industrial agricultural practices, e.g. killing plants with chemical herbicides and then eating them. So naturally, my first instinct was to wonder: did they approve a weird new pesticide for use on peanuts in 1980? Maybe eating peanuts laced with some chemical toxin could train the body into thinking they're bad business.

I did a quick search along those lines—something like peanut agricultural approval 1980. I didn't find anything like what I was expecting. Instead, I got a slew of hits in the form of old newspaper articles, all reporting on something called the “Great Peanut Crisis of 1980”.

Sounds like a lead. Was it a crisis because that was the year people started dying from eating peanuts?

Nope! Turns out the peanut crisis was mostly an economic one: the result of a drought that hit the Southern US that summer, killing about half the crop and causing prices to triple. To make matters worse, the dead and dying plants became a breeding ground for an agricultural pathogen—a fungus called Aspergillus flavus—so something like 15% of the surviving ones ended up too moldy to eat.

And suddenly everything started to click into place, because Aspergillus flavus is known for producing one of the absolute gnarliest chemicals in existence: aflatoxin.

I first learned about aflatoxin back in undergrad, in a Radiation Biophysics class. The professor used it as an example: some certain dose of x-rays (maybe a standard dental x-ray, I don’t remember) would increase your risk of cancer by about the same amount as eating a jar of peanut butter. The vast majority of jars are fine, of course, but every now and then a moldy batch of peanuts sneaks through the quality control process. If you get one that had a few aflatoxin-tainted peanuts in it, that’s often enough to give a person liver cancer.

And if that seems as weird to you as it did to me then, consider: the x-rays don’t want to hurt you. They’re just very high-energy, and as they stream through your body they’re bound to knock a few bits loose from your DNA. And while there’s a chance this indifferent damage can break one of the genes that tells a cell when to stop growing—causing it to become cancerous—you have to be pretty unlucky.

Aflatoxin, by comparison, is malice incarnate. It distinctly wishes you ill. Its creator, Aspergillus flavus, does not want to be digested. Nor does it have any interest in joining the human gut microbiome. It would rather do anything besides help you ferment rice like its cousin Aspergillus oryzae. It wants to be left alone with its peanuts, thank you very much. And so it produces one of the most carcinogenic substances on Earth. A chemical weapon honed by a million years of evolution into a shape that can effortlessly pop bases out of your DNA, causing your liver cells to go haywire in any number of horrific ways.

We typically think of cancer initiation as something that happens quietly, without our noticing (Will this be the cigarette that does it? There’s no way to know…) but aflatoxin is potent enough to often make the victim acutely ill—usually with a kind of hepatitis—as random gene mutations cause some cells to grow distended and misshapen, while others simply die en masse. Aflatoxin wants you to feel the hurt, because its whole purpose, its raison d’etre, the thing that gets it out of bed in the morning, is to make sure that—if you ever see a moldy peanut and think “eh, probably fine to eat?”—you never make that mistake again.

So it would not surprise me if aflatoxin were capable of causing the immune system to do some buck-wild things, e.g. developing such a strong response against a harmless food protein that—the next time you’re exposed to even a crumb of that food—you literally die of anaphylactic shock. And while I don’t know much about immunology, this model jibes with my understanding of how things like attenuated vaccines work: you inject somebody with a bit of pathogen which, on its own, wouldn’t trigger much of an immune response, because the thing is practically Dead on Arrival. So you throw in an adjuvant, a chemical that induces a local inflammatory response—usually some kind of aluminum salt, these days. Immune cells pick up the distress signal, the metaphorical cops arrive to the scene and find a bunch of broken windows and this one unconscious ruffian, they come to the obvious conclusion, and load him into the back of the wagon hollering about how he’s been set up.

Here, I guess, the role of “pathogen” is played by some peanut protein sturdy enough to survive the stomach intact,3 the “adjuvant” is the aflatoxin, and instead of a few “broken windows”, the crime scene is so spectacularly gory that, next time there’s a confirmed sighting of the suspect, the president doesn’t hesitate to call down the nukes on his own soil.

To my mind, this is consistent with the narrative that’s been taking shape over the past few years, in the form of books like Blind Spots by Marty Makary. The argument—backed up by some pretty solid interventional trials—is that we’ve been making the problem worse by giving in to fear: that deliberate, early, and extensive exposure can prevent a lot of peanut allergy. And yeah, the body is pretty good at discerning friendly from unfriendly molecular signatures as long as it has enough training data; that’s how your immune system knows not to attack your own cells. So you’d expect the problem to be worst in people who had little or no exposure to peanuts before getting a tainted batch of butter, and whose parents then decided to play it safe and switch to almonds.

The Obvious Thing

Could it be that simple? That obvious—that the first few hits when you google “peanut agriculture 1980” contain all the information we need to make sense of this societal-scale medical mystery? Anyone who says “no” has lost sight of the uniqueness of the present moment in time, where all the world’s knowledge is suddenly at our fingertips.

I asked about it at that lab meeting. Was anyone looking into this? The institute director waved it away; “Aflatoxin is a problem of heat and humidity and poor food storage. In the developed world, exposure has probably been trending downward for all of human history.”

I put my hand down. It sounded sensible enough, and I had learned not to argue with Cezmi.

Two seconds later, I was kicking myself silently: Wait, no, it’s the peanut butter! That’s your whole shtick, isn’t it? A uniquely American problem, because everywhere else on Earth, if you crack open a peanut shell and see mold, you don’t fucking eat it!

And this is what makes it a compelling fit with the epidemiology, to me—with the fact that peanuts never killed anyone before 1980, and still rarely do except in the USA.4 I mean, put yourself in the shoes of a farmer living through that crisis year. Half your usual crop never materialized, and half of what’s left is somewhere between “spotty” and “lousy” with Aspergillus. Imagine you were in a tough spot financially, even before the drought. And imagine that you’re genuinely ignorant about how bad aflatoxin can be for the people exposed to it.

Might you not try and find a way to recoup a little, in this disaster of a year? Unload the more passable fraction of the spoiled ones, whether or not they’d technically pass muster with a USDA inspector (who almost certainly has his hands full this year anyway)? They’re paying triple what they usually do.

And what better way to take a half-bad batch and make it saleable than by turning it into a homogeneous, undifferentiated mass? What better way to cover your tracks than…peanut butter?

It seems right, in a way, that a societal ill as weird and capriciously deadly as peanut allergy should be a consequence of an equally weird sin. “Bundling toxic assets”; refusing to eat the loss when fate deals you an unlucky hand. Imagining that, if you spread it around, maybe you can thin it out enough that it doesn’t do anyone real harm.

But of course it doesn’t work like that. Maybe it would, if you just had to fool the eye: a quick trip through the blender and, to someone looking at it, there’s none of the instinctive revulsion that you get from the thought of eating a green peanut. But that instinct is there for a very good reason: because one of your ancestors dealt with an Aspergillus or something very much like it—and the experience left them sweating and retching and writhing in such agony that, countless generations later, you can still feel the echo of their misery: a chill down your spine when you unwrap the tinfoil around a mysterious hunk at the back of your fridge to find colorful fuzz.

A lot of times in posts like this, I’ll finish by talking about how you’d test the hypothesis, but I’m not even sure how you would go about investigating this one. Maybe kids with peanut allergies would also have antibodies against aflatoxin in their system at a higher rate than controls. Then again, maybe not: part of what makes aluminum sulfate a good adjuvant is that, for whatever reason, the immune system has a hard time recognizing it. It wouldn’t surprise me if aflatoxin were similarly slippery, especially given that it’s tailor-made to mutate DNA and disrupt cellular functions. You might be able to test for the toxin itself in individual jars, if you catch it the first time a kid has a bad reaction to peanut butter.

But that brings us to the practical takeaway: if this hypothesis is correct, it shouldn’t be that hard to give your kid a pretty good shot at avoiding the ignominy of having “The Goobers Got’em” on their tombstone. Just start feeding them peanuts—actual whole peanuts—early. Because the core conceit here is that a lot of the problem with “processed” foods has less to do with the processing, and more that you don’t get to really see the food before you eat it, and so you end up eating things you wouldn’t otherwise: rancid oil, spoiled grains, etc.

So if you want peanut butter—at least for the first couple of years, while your kid’s immune system is still figuring things out—make it at home. Buy roasted peanuts at the store and chuck them in the blender. But have a look at them first. You probably don’t need to—ones slated to go to the consumer whole are held to much higher QC standards than ones destined for processing…but it can’t hurt to pick out any weird ones.

—🖖🏻💩

Share

Tremendous thanks to Gautam Panakkal for encouraging me to get this one out.

I realized that something did not add up when I was about ten years old. Went to the museum during day-camp one summer. Saw a real human skeleton for the first time—a pair, actually. One male and one female.

“What differences do we notice between the two?” the counselor asked.

I made a quick count. My hand shot up, then I spoke without waiting to be called on. I was that kind of kid.

“The female’s has one more set of ribs” I said.

She frowned. looked over at the skeletons.

“No-o-o,” she said, “but great bible knowledge!”

Mortifying! I had miscounted—missed a floating rib on the male or something—and there’s nothing worse to a know-it-all kid than being so publicly, obviously wrong.

But looking back, I’m proud of young Stephen for making that assumption. in first grade, our teacher had read us a folk tale about why Anansi the spider has long legs, and even at that young age I understood: these stories are for explaining the world.

Anthropologists call them Just-So stories: Hephaestus thunders away in his forge, hard at work on Zeus’s next lightning bolt. Pyramus and Thisbe bleed out under a mulberry tree, and the gods of Babylon enchant the tree such that its fruits—which sprout milky white from the branch—ripen to a bloody red-black forevermore.

Genesis

So in what WEAK-ASS cosmology does your supreme deity take a bone from the prototype of your species, and then the next generation it just...what, grows back? Imagine someone tried to tell you about the time God took the stripe from the skunk! Such a story could only be entertained in anticipation of the next installment, “How Skunk Got Her Stripe Back”, right?

So why, in the first chapter of the first book of the foundational text of the most popular religions on Earth, is there a Just-So story about “why man is missing a rib”?

I suppose if you earnestly believe “Oh all those other stories that follow the same form, THOSE are myths, but this is a true, literal account of how God created women”, then it doesn't have to make sense through an anthropologist’s lens. But if you see the Abrahamic creation myth as being fundamentally no different from the Babylonians’, or the Hopi’s, it is so obviously a just-so story about “why man is missing a rib” that my ten-year-old self felt comfortable making assumptions about human anatomy based on it.

What gives?

Well, a few years ago I got curious: does the original text really specify a “rib”, like from the ribcage? After all, YHWH did not hand the Commandments down to Moses on stone tablets written in the King’s English; the version of the bible that most of us are familiar with is the product of a millennia-long game of Telephone. Theologians have been grappling with the implications of this since the Coptic Bibles were written, because if you want to seriously claim that your text is the Sacred and Incorruptible Word of God, you find yourself asserting that God spent a lot of the fourth century peering over the shoulders of scribe monks as they hand-copied manuscripts going “sorry, that’s an i, not a j”.

No trouble for the omnipresent Almighty, to be sure, but very out of character after the whole Tower of Babel thing.

Fortunately, we still have the Hebrew—the closest we’re ever gonna get to the original text as-written—and Strong’s Concordance, an exhaustive scholarly effort which indexes each time a word is used in the book, so you can compare and use context clues to deduce what the ancients might have meant.

And if you check Strong’s…

Aha! There’s apparently some flex in the translation of the word commonly taken to mean “rib”. It can refer to a wooden plank—any kind of structural beam, really. Also worth noting: it’s very clearly not a set of ribs, or a pair; it’s one bone. If it came from the ribcage, this would seem to be a just-so story about why humans have a slight asymmetry. But the thing that really caught my attention were the final two words of the sentence. תַּתַּחְתֶּֽנָּה׃ (taḥ·ten·nāh) comes from תַּחַת, tachath, which means “below” or “under” in practically every context where it’s used. And the final word, בָּשָׂ֖ר (bā·śār)…

There’s a reading of Genesis 2:21 that goes: “…and sealed him up underneath the dick.”

And the moment I saw those words, three memories collided in my mind.

The first was of my embarrassment at summer camp all those years ago; my confusion.

The second, a memory of being a very young boy, of examining my own body and really noticing this suture-like seam between my legs for the first time. Bending over and looking in the bathroom mirror at it, wondering. Imagining for a minute that maybe I was born a hermaphrodite and they had sewn me up.

The third memory was of a bit of anatomical trivia.

See, the human male is missing a bone—not compared to the female of his species, but relative to the males of most other species of animal.

We alone among our closest cousins, the great apes, are the only ones without a baculum.

A penis bone.

Boneless Tenders

And as these recollections flashed before my eyes, I was struck with an almost religious certainty. I could see the ancestors, butchering a bird or a bear, going “huh, these guys have one too.” Looking down, wondering what makes us different, speculating.

I went “has anyone else thought of this?” and of course, there’s nothing new under the sun; I found an article from the American Journal of Medical Genetics, Gilbert & Zevit (2001)—postulating this very same thing, on basically the same line of evidence.

Thus I find myself a Zevitist—or maybe a “Baculist” is more coalitional—on the question of the biblical rib.

It’s frankly astonishing to me that, even though the slightly-mistranslated and censored version of this story is so widely known, it’s so rarely given enough thought for this to be a popular opinion. But the fact that I am not the only one to have seen through the words written—to the story as it was originally told—gives me hope. Rabbi Zevit and I both had the same insight independently, which assures me: we are part of a long tradition. As long as the story is still told in some way, there’ll be a few people in every generation who think about it until they get it, who go back to the old texts, who join this club that has no home in time or space, whose motto is: “They’ll never believe you.”

It’s too silly! Makes it seem too much like what it is: a prehistoric joke—one that caused such a ruckus around the campfire that it managed to get repeated and remembered until writing was invented.

But I think, like all good jokes, there’s a grain of truth at its heart. It’s still a good question: Why don’t humans have a baculum?

Scientists have speculated endlessly about the evolutionary pressure that led to “congenital human baculum deficiency”, as Gilbert & Zevit put it. It’s an intriguing puzzle, because when you see a trait present in every member of a species, and absent among every phylogenetically adjacent species, there’s a decent chance you’re looking at a speciating trait—the mutational fork at which two branches of the tree of life diverged. But baculum loss seems like it’d hurt your chances of passing on your genes, since even today it renders the human penis more susceptible to catastrophic breakage. (On Mother Nature’s cutting room floor, next to the heaped bodies of baby giraffes whose necks were too weak to handle the six-foot-drop that is the first thing one experiences upon being born as a giraffe, lies the pile of dudes who broke their dicks attempting a 720° Reverse Cowgirl Frontside Bluntslide. We weep for the giraffes, anyway.)1

In The Selfish Gene, Dawkins suggests that baculum loss is a bit of “honest signaling” like the peacock’s flashy tail—replacing the efficient baculum with an Inflatable Arm-Flailing Tube Man powered by blood pressure, to prove you’ve got the cardiovascular chops to sire healthy kids. This is not particularly compelling to me (in part because “it’s deliberately inefficient” can be used to wave away practically anything you can’t explain) and the theories only get worse from there: Go to the Wikipedia page for “baculum” and, as of this writing, you can still see someone seriously suggesting that proto-humans learned to break each other’s dick-bones in combat, and this rapidly extirpated those with a bone there to break.

But that’s generally not how evolution works: Evolution happens one freak, one mutation, at a time. At some point, a proto-human ape was born without a baculum, and our entire species as it exists today is descended from that first man. But if not because he, alone, could asymmetrically deploy the proto-human dick-bone kung-fu snap—then why? How could this crippling fragility be a fitness advantage?

It’s obvious, of course, if you put yourself in his shoes. You’re a mutant—although you might not know it. You’re probably not talking about it with your guy friends; maybe we don’t even have real language, yet. Maybe we’re still furry.

But for you, sex is not like for the rest of them. Unique among your species, you have to rely on some flimsy erectile tissue, effectively a chilly nipple. The other guys can just prop up the tent pole and go to town, while the best you can manage is some…engorgement.

In other words, you are the first guy in history who has to be very, very careful when putting it in.

You can’t even do so, really, without the full enthusiasm and assistance of your partner.

Without consent.

Baculum loss turns sex from what it is in the rest of the animal kingdom—“nasty, brutish, and short”—into something collaborative. Something requiring a lot of communication. Something, dare I say it? Tender.

So imagine what a sensation—what a revolution—this Adam must have been with the ladies! His first partner, perhaps confused by his difference, then amazed to discover: there’s a better way. It can be gentle. Pleasant, even.

You bet your ass she’s telling her friends about it! Words? Who needs ‘em! She’s dragging them over for show-and-tell, ook-ing and pointing, squeezing his junk to demonstrate, gesturing for them to do the same: No, feel! This one’s built different!

And I imagine our man, who never knew he was anything special, is probably thinking to himself something like: Okay, not sure what all this is about, but it seems like a good thing, right? Let’s play it cool, let’s see where this goes…and in a few generations, Adam is the common ancestor of all males in his local group; a floppy Genghis Khan who won supremacy not with strength, but with vulnerability.

Think of the balance of power between the sexes in a proto-human society, and how it would shift in such a tribe!

Who is the alpha? Who gets to breed this season? It’s not for the men to decide among themselves anymore; it’s each female’s choice.

Would we not expect that group to have an unusual degree of intimacy between partners? Unprecedented cooperativity among males, freed from the alpha/beta framework? With the monopoly of brute-strength-as-reproductive-fitness broken by this newfound physical fragility—giving females veto power over copulation—would we not expect an explosion of genetic diversity, as multiple males per generation get a chance to mate, competing on a playing field of the females’ devising? Might such a tribe not start selecting for things like cleverness, empathy, and the capacity to communicate?

Might that group not take over the whole goddamn world?

So maybe this is the grain of truth. Not as separate acts, but all in one fell swoop, God made man, and took the bone, and made woman his partner.

—🖖🏼💩

P.S.
I would be remiss not to discuss the other exceptional component of human reproductive anatomy: we have the largest penis-to-body-size ratio of any ape. For comparison, a well-endowed gorilla clocks in at somewhere around 2 inches.2

Why?

Here again, theories abound, but the true answer is self-evident once you hear it: If you want to understand the shape of a key, you have to look at the lock it was cut to fit. So we reframe the question: it’s not “Why do humans have such disproportionately large and girthy members?”, it’s “Why do humans have such disproportionately large vaginas?”

And the answer to that one is obvious: during birth, they have to accommodate our disproportionately large heads! As we entered the feedback loop of larger brains enabling higher cognition enabling the acquisition of more and better food, the human vagina—and by extension the penis—had to grow in parallel to the cranium. Perhaps the ultimate irony, then, is that we would eventually reach a point where male genitals were large and sturdy enough to enable the reinvention of rape.3 The stretch of history between baculum loss and this evolutionary threshold is, presumably, what we refer to when we talk about “Eden”: the days before Adam succumbed to Eve’s temptation and discovered what it means to be evil.

“Sulfonylurea” is a word you're probably only familiar with if you’re:
A) a farmer, or
B) a doctor who was practicing medicine 40 years ago.

Old doctors know about sulfonylureas because they were some of the first drugs developed to treat diabetes.1

Farmers know about sulfonylureas because they’re the most heavily used herbicides in the world today.

Maybe you already see the problem.

The Farmers

There are at least two dozen different chemicals on the market with the general structure shown above, and they're used on practically everything: corn, wheat, rice, soy, sugarcane, etc.

Sulfonylurea herbicides were invented in the mid-’70s, and came into use starting in the early ‘80s. Usage ramped up slowly at first, then exponentially: by 2011, sulfonylurea use had hit more than 2 million kg/year worldwide, and it’s only grown since then.

Their introduction transformed agriculture to a degree that’s hard to overstate, because—in addition to being about a hundred times more potent than herbicides in use at the time—they were also substantially more selective.

See, in the old days, weedkillers were used in commercial agriculture in much the same way you’d use them in your garden at home: walking around with a spray tank and spot-treating any interlopers you spot among the rows of crops. Sulfonylureas were a revolution because many food crops have some natural resistance to them: enzymes that metabolize the chemical once it’s absorbed, breaking it down into compounds that don’t have the same herbicidal effect. Most weeds, on the other hand, don’t have those enzymes—so they just up and die.

Suddenly, with the invention of sulfonylureas, we went from having to spot-treat weeds to being able to spray down the entire field and kill just the weeds, sparing the crops.

And the reason this was such a game changer is that walking around with a tank and sprayer, blasting weeds one by one, isn’t that much less work than pulling the weeds by hand. It’s also more expensive, since you have to buy the chemicals. That was the other key advantage of sulfonylureas: when a few milligrams of the stuff is all it takes to kill a plant, you can afford to spray the whole acre, even if 90% of it is going to waste by landing on plants that it won’t kill.2

All this set the mold for “no-till” agriculture, i.e. relying on chemical rather than physical means to keep fields weed-free. Since then, this style of farming has pretty much taken over the world, because it means one person with the right equipment can do the work that previously might have taken fifty. Who wants to traipse around pulling or spraying weeds by hand when you can just drag a 30-row sprayer behind your tractor and be done with the field before lunch?

If you read my piece on Roundup, AKA glyphosate, maybe you’ve noticed the parallels in how sulfonylureas and glyphosate are used. That’s no coincidence: the development of roundup was, in a lot of ways, an effort to replicate the zero-to-one success story of sulfonylureas. The two even have similar mechanisms of action: they kill plants by inhibiting the production of certain amino acids.

Sulfonylureas work by shutting down an enzyme called acetolactate synthase, preventing the synthesis of the branched chain amino acids—things like valine and leucine. Glyphosate inhibits an enzyme called EPSP synthase, which is essential for the production of the aromatic amino acids, e.g. tryptophan and tyrosine.

In practice, the only major difference in how they work is that, where many food plants are naturally resistant to sulfonylureas, practically the only things naturally resistant to glyphosate are:

1. Organisms that don't make aromatic amino acids on their own, like Homo sapiens, and

2. A weird strain of Agrobacterium that the guys at Monsanto fished out of the drain, which can make aromatic amino acids even in the presence of glyphosate because it has a mutant version of the enzyme that the herbicide targets.

Hack that mutant gene into your crops, and you’ve effectively recreated the sulfonylurea paradigm: spray the whole field with roundup and only the weeds die, leaving behind eerily pristine rows of corn. The debut of “roundup ready” crops in the late ‘90s was an especially big deal to everyone who had already converted to no-till farming, because sulfonylurea-resistant weeds were already becoming a huge problem. It was a big deal for Monsanto, too: now, not only did they get to sell two different classes of herbicide3, they also got to sell the seeds that the new one could be used on safely.

“Safely”, in this case, of course means safe for the plant, for yields, and for the farmer’s bottom line. But for the people eating those plants?

And this is the crux of my issue with glyphosate, sulfonylureas, and “no-till” agriculture in general: Unlike insecticides or fungicides, which only have to sit on the surface of the plant to work (and thus can theoretically be washed off), an herbicide can’t kill weeds unless the weeds absorb it. This means that, when you spray a whole field of food crops with poison, you necessarily end up putting poison in the food at concentrations sufficient to harm living things. So you don’t do that unless you’ve got some clever reason why the poison won’t harm you, right?

With Roundup, it’s that we’re effectively naturally immune to it—we don't have the enzyme glyphosate targets. Unfortunately, plenty of our gut bacteria do.4

But that's not the problem with sulfonylureas. Even though many of your gut bacteria also make branched-chain amino acids (and thus rely on acetolactate synthase), remember that crops protect themselves from the toxic effects of sulfonylureas by breaking them down, detoxifying them—so they ought to be inactivated well before they enter your body, or else they’d have killed the plant.

So what’s the problem?

The problem is that, although the resulting metabolites don't have the same herbicidal activity as their “parent” compounds, they’re not necessarily inert. To the extent that humans are different from plants (which is to say: quite a lot, at least for most people) a chemical that’s harmless to a plant might be bad for a human in weird ways. And in the specific case of sulfonylureas, there are some very good reasons to suspect that this is so.

The Doctors

See, well before they found use in agriculture, sulfonylureas were pharmaceuticals. Back in the ‘60s, drugs like tolbutamide were among the first small-molecule therapies for diabetes. They work by triggering the pancreas to squirt out insulin, which lowers your blood sugar.

But sulfonylureas aren’t used much in medicine anymore, because—while it’s helpful in the short term to be able to mash the “release insulin” button just by taking a pill—artificially yo-yoing your insulin levels over long periods of time can have some negative consequences. If you take them constantly, you can get tolerance. Insulin resistance.

And this is worrisome, because the fundamental problem in diabetes, whether it’s type 1 or type 2, is that the calories in the food you eat aren’t getting turned into energy the way they should. The primary symptom is high blood sugar, but this isn’t just a consequence of eating too much sugar—it’s that the sugar doesn’t have anywhere to go, thanks to a breakdown in insulin signaling.

See, your pancreas’s job is to monitor the level of sugar in your bloodstream, and when it rises, release insulin—the chemical signal that passes the message to your cells: “Hey, there’s sugar in the bloodstream; you should take that up and put it to use.” Muscle cells use that glucose for energy, breaking it down to produce ATP, while fat cells convert it into lipids and bank it up.

In type 1 diabetes, the problem is that—for whatever reason—your pancreas is busted and can’t make insulin, so the blood sugar signal never gets sent. This is why injectable insulin, the first major drug developed for diabetes, works like gangbusters in type 1: it effectively solves the core problem, as long as you remember to take your shots with meals.

But type 2 diabetes is a different beast, and an altogether more complicated one. Here, the pancreas releases insulin in response to meals, but the insulin doesn’t seem to work as well. This is the key feature of type 2: insulin resistance—meaning the cells of your muscles, liver, etc. don’t take up as much blood sugar as they should in response to a given amount of insulin. So sugar just hangs out in the bloodstream, and your cells—basically unaware that there’s an abundance of calories just waiting to be had—go into low-power mode. Convinced that you’re starving at a cellular level, you end up in this paradoxical state of too many calories and not enough energy.

The popular conception is that this works something like a drug-tolerance effect. If you drink enough coffee for long enough, your body learns to ignore it. Your neurons express more adenosine receptors, dampening the effects of the caffeine until eventually you can enjoy a nice cuppa at bedtime without hurting the quality of your sleep. Likewise, the thinking goes, if you eat too much sugar for too long, you get insulin resistance.5

And the same way the caffeine addict can still get a boost from coffee the next day (provided the barista doesn’t start reaching for the panic button under the counter when you say “octuple espresso, please”), people with type 2 diabetes can still benefit from treatment with insulin…but it’s not a solution-to-the-problem kind of benefit; it’s just one tool in the toolbox for managing the disease.

But the insulin-resistance-as-sugar-tolerance analogy has some assumptions baked in that don’t sit quite right. Glucose isn’t a drug like caffeine; it’s one of the fundamental currencies of energy in biology. Putting it to use properly is as essential as processes in the human body can get.

So how do we enter the maladaptive state of insulin resistance?

The conventional answer

…is “Too many calories in, not enough calories out”—and it’s true that we live in a time of unprecedented caloric abundance and hyper-palatable food. As a society, we’re sedentary in a way historically reserved for lounging emperors and double amputees.

You could argue that sustained excess like this is such an evolutionary edge-case that the body can’t handle it. Physical inactivity is one of the main lifestyle factors associated with insulin resistance and diabetes. So maybe, the thinking goes, your cells factor in the demand on them when establishing their set-point for how much sugar to take up from the bloodstream. After a while under low demand they lose the sense of what’s optimal.

But this would require the human body to be an uncharacteristically shoddy piece of work on evolution’s part, because in someone with T2D, the cells clearly aren’t meeting the demands on them with the amount of sugar they take up. You can tell, because starting somebody with type 2 diabetes on insulin often yields huge improvements in their energy levels—both physical and mental.

So while calorie-dense foods and physical inactivity are tightly correlated with insulin resistance and diabetes, we should be wary of the intuition-trap that correlations like this can present. After all: if you snuck into someone’s house and randomly slipped drugs into half their food which scrambled the signals that let their body turn that food into energy…would you be surprised to find them overeating, or having a hard time getting up off the couch?

Because that is effectively what’s been happening to us.

Caveats & Gremlins

Some things I should clarify:

1. “Sulfonylurea” is a broad chemical class, and just because two compounds are in the same chemical class doesn’t necessarily mean they’ll have the same effects on the body. Structure is function, both in chemistry and biology, but the ways in which that fact manifests are often hard to predict; the addition or subtraction of just a few atoms can magnify, eliminate, or even invert a chemical’s effect. To give a popular example, “substituted phenethylamines” is a class that includes everything from sudafed to MDMA to potent psychedelics like 2C-P. All three will keep you awake, and at least two will clear your sinuses, but only that last one will melt your psyche into the shadow of a stormcloud on the sea.

2. None of the sulfonylureas used as herbicides are identical to the ones used as diabetes drugs. And, given point 1 above, they might not have the same insulin-releasing activity as their pharmaceutical cousins.

So we can’t say for sure that agricultural use of sulfonylureas ought to mess with insulin signaling.

What we can say for sure is that we don’t know whether the agricultural use of sulfonylureas messes with insulin signaling.

I can hear you asking “what do you mean, we don’t know? Surely these chemicals were tested for safety before being licensed for application to food crops. They'd have noticed if it gave a bunch of mice diabetes.”

And I imagine they were—although I have dug pretty deep into the literature, and haven’t found any reports of such tests. The closest thing I can find is this case series from Greece, where some doctors noticed the weird coincidence of T2D in three patients with healthy BMIs, no risk factors, and no family history of diabetes…all three of whom were farmers that used sulfonylureas in their work.

But the reason I can say with certainty that we don’t know whether agricultural use of sulfonylureas messes with insulin signaling is that we’re not eating the chemicals themselves. We’re eating the plant metabolites—the breakdown products—and in many cases, we still don’t know what those metabolites are.

See, the enzymes responsible for the metabolism of sulfonylureas mostly belong to a family called the cytochrome P450s, or CYPs. This is an extraordinarily diverse class of enzyme—all variations on a gene that probably existed in the Last Universal Common Ancestor, as evidenced by the fact that they can be found in everything from bacteria to plants to humans. And although they mostly serve the same general purpose—helping their respective owners degrade and eliminate toxins by sticking an oxygen onto them—each organism has its own versions, which bind different molecules and modify them at different spots. Humans, for instance, have some 57 different versions of CYP in our genomes, with names like CYP1A2 (which metabolizes tylenol but also turns certain unsaturated fatty acids into immune-signaling molecules), or CYP3A4 (breaks down most benzos, but is famously susceptible to blockage by chemicals in certain citrus fruits), or CYP2D6 (deactivates SSRIs and amphetamines, but also transforms codeine into morphine, enhancing its potency by about 10x). To further complicate things, chemical cues can induce a cell to produce more or less of a given CYP enzyme. This induction is used to great effect in agriculture, by combining sulfonylureas with other chemicals referred to as safeners, which cause the crop to ramp up production of the appropriate CYP enzyme.

So maybe it’s not so hard to believe that we don’t even have a list of all the chemicals that might end up in your body as a result of sulfonylureas’ use on food crops (much less an understanding of those chemicals’ effects on our biology), is it? Consider what a massive undertaking it would be just to identify them: Take each of the dozens of chemicals listed in the screenshot at the top of this post and multiply it by the number of food crops it’s approved for use on (ideally, the number of varietials, since apparently sweet corn and feed corn have different enough CYPs to change their profiles). Multiply that by the number of safeners approved for that [crop] x [herbicide] combo. Now run that many dedicated metabolomics projects to figure out what each of those chemicals turns into when the plants metabolize them.

From the handful of [crop] x [herbicide] combos where that’s been done, we know that one sulfonylurea can turn into three or four different metabolites in a given plant—so your full list would likely be several hundred compounds long, if not several thousand. Now, synthesize or purify a few grams of each of those compounds and feed it to a few cages of mice at the relevant doses for three months to figure out which ones cause problems. Oh, but first you’ve got to invent USDA-organic rodent chow, because otherwise your control group will be eating a random smattering of your test compounds. And really, if you’re going to all that trouble, you should check which of those plant CYP metabolites are subject to further metabolism by human CYPs, because those aren’t the same as mouse CYPs. So you’ll want to add all those metabolites to the mouse test list, and figure out some way to prevent the mouse CYPs from doing things that the human versions wouldn’t.

It would take a Manhattan Project.

Earlier, I threw out a number: 2 million kg of sulfonylureas got used as herbicides in 2011. That’s about 250mg for every person on Earth at the time.

Now consider that, when sulfonylureas are used as diabetes drugs, the newer ones like glimepiride are dosed in the range of 1-2 milligrams. That’s how much it takes to make a major impact on a person’s insulin signaling.

In the subtitle of the post I said “this is good news”, because it almost seems too good to be true, as far as theories go. From our vantage point in the present moment, where there's a Wikipedia page on sulfonylureas that has an “Agricultural Use” subheading right next to the “Pharmaceutical Use” one, the idea seems so colossally obvious that to have missed it ‘til now would require implausible levels of incompetence on the part of everyone involved. It’s easy to forget that—until very recently—information like this has been siloed to such an extent that, by the time someone with the requisite knowledge to flag sulfonylurea herbicides as “potentially a really bad idea?” got wind of it, the project could already have had too much inertia to be halted by a hypothetical safety concern.

But something big is responsible for the absolute pandemic of metabolic dysfunction that’s hit the USA over the past 50 years. It tracks pretty well with the rise of cheap, processed foods. But “processed foods” is such an aggregate proxy. The fraction of an average person’s daily calories that comes from ingredients like “enriched, bleached wheat flour”. Is it the enriching? The bleaching? The seed oils, the soy lecithin, the corn syrup?

But step back, and follow the money: why did corn syrup suddenly get so cheap that Coca Cola was willing to pull elaborate hijinks to smooth over the recipe transition from cane sugar to HFCS?

The answer is the agriculture, the new no-till paradigm—and the first big break in that direction was the development of sulfonylurea herbicides.

So it does not seem unreasonable to ask: What if sugar itself is not the culprit here? What if it has something to do with the analogues of known insulinergic drugs that we recently started spraying on the crops?

Call me crazy. Or call your congressman.

Because given:
1) the difficulty involved in answering that question in the lab,
2) the scope and stakes of the current obesity crisis, and
3) the plausibility of the notion that it could have a simple, singular cause…

the reasonable thing to do is to rescind the approvals for these chemicals, at least for use on food and feed. Phase Them Out; that’s the slogan, nice and chantable. If it’s done right, we will know within a few years whether or not that was the problem. And hey, if there’s no improvement in the average American’s metabolic health, phase ‘em back in!

This will not be simple, no matter how simple the idea behind it is. It will require advance notice, to allow for a massive restructuring of the agriculture industry, if we don’t want to just replace sulfonylureas with new and different poisons. It will take subsidies and jobs programs to muster the manpower to grow the food without them. It will take amnesty clauses for all parties involved, to dissolve industry resistance. It will take bailouts, paying the chemical companies for material they don't produce. This is fine, or at least no more ridiculous than the current scheme of paying farmers for grain they don't plant. It will require taking a leaf out of China’s book, turning away shipments of sugar that test positive for prohibited compounds.

But it can be done. We’ve currently got grain in such abundance that we can take the hit to productivity without anyone starving. The political will is there: The president’s palanquin was carried to victory this last election in part by people like RFK Jr. and Casey Means—the perpetual nominee for surgeon general—whose whole shtick is specifically that our highest priority should be figuring out what is going on with metabolic syndrome. And we have the technology: In California, the combination of gas taxes and EV credits was enough to turn Teslas from a fringe phenomenon into a substantial fraction of vehicles on the road; we can apply the same pressures to convert the nation’s sprayer-tractors into laser-tractors, computer-vision equipped to fight the war on weeds with greater tactical precision than we ever could have achieved with crude chemical means.

Because if not for this, then for what?

The process of directing energy—taking it in as calories and turning it into useful work—is as fundamental as it gets. It is pretty much the whole business of being a living thing. If we cannot be bothered to figure out what is going wrong with our ability to do that—if we do not have what it takes to run this Great American Experiment as a proper experiment, with a crossover control arm—for the sake of our people’s capacity to be strong in arm and sharp of mind?

Then folks, we as a society are COOKED.

Phase ‘em out!

—🖖🏼💩

Imagine how completely the experience would rock his world: the richness, the mouthfeel—unlike anything he’s ever known; he’s barely got the hang of cooking meat over a fire. Imagine the look on his face as he tries to comprehend the aroma, luxurious and enticing but somehow also reminiscent of…feet?

Now imagine someone did the same thing to you.

Imagine some time-traveler came and brought you a food as gobsmacking—as earth-shattering—as the brie would be to the caveman. Something equal parts strange and wonderful: a texture that you didn’t know was possible, a new and exotic funk that even the most well-equipped flavor chemists in the world couldn't dream up.1

Spectacular new foods like that are possible. Maybe even coming soon…if we can stop chasing chickens.

Hubris

These days, a lot of the conversation around “the future of food” centers on things like lab-grown meat. Startups like Upside Foods—working toward the vision of a cruelty-free steak made from animal cells grown in a tank—have raised eye-watering sums from investors. But there’s a reason you’re not seeing it on shelves yet: without massive breakthroughs in tissue culture technology, there's practically no way for lab-grown meats to be commercially viable anytime soon.

But there is a better way. An easier way: One that's both more likely to succeed, and infinitely more rewarding once it does.

After all, even if those technological breakthroughs come tomorrow, and lab-grown meat becomes indistinguishable from the real thing…what have we really gained? Slightly cheaper chicken?2 Tissue cells grown in culture still need food, so—perhaps ironically—a lab-grown chicken nugget will probably be raised on the same old soybeans as the regular kind…once those beans have been defatted, hydrolyzed, and processed into powder.

It's still a noble goal, to be sure—the effort to transform low-quality calorie sources into tastier and more nutritious ones is the overarching theme of humanity's entire agricultural odyssey—but the approach here is too clever by half.

Think about humanity's greatest culinary achievements—the ones that literally shape the modern world: Cheese. Chocolate. Wine. Did we give up the easygoing hunter-gatherer lifestyle for grain, for crackers? No! It was only when bread—and, even better—beer came into the equation that it suddenly became a good bargain for man to put down his atlatl and pick up the plow.3

And the thing that all these foods have in common is that they are products of fermentation, which—at its heart—is the same kind of caloric upcycling that the scientists at Upside and other lab-grown meat companies are trying to pull off. But these foods don't need pressurized cleanrooms and gamma-ray-sterilized equipment. They can be made with stone-age technology, because—rather than playing God, trying to outdo nature—they’re the result of discovering nature's readymade gifts and embracing them.

Look back across history: most of the truly revolutionary advances in food science have come from people discovering how to harness the power of microbes. And looking forward, now that modern science has given us a glimpse of just how rich the microscopic world really is, it’s clear that we’ve only just scratched the surface of what’s possible.

The State of the Art

Nearly all existing fermented foods are produced by just a few different kinds of microbes. Look into the bacterial makeup of fermented foods from various cultures around the world, and you might be surprised to see the same names coming up over and over again: Japan’s famous natto, Nigeria’s Kpaye, even the coffee and cocoa beans all owe some of their unique properties to the action of Bacillus subtilis. The yeast Saccharomyces is almost singlehandedly responsible for all alcoholic beverages,4 while the various Lactobacillales give us cheese, yogurt, kimchi, pickles, sauerkraut—even prosciutto. With an assist from Propionibacterium, or Streptococcus, a Lacto can make kefir or Swiss cheese. In concert with Saccharomyces, it can make a sour beer.

But this incredible variety of products hides the fact that the organisms which help produce them represent only a select few branches on the tree of life: the useful microbes which we lucked into discovering before we even understood what they were. It’s a pretty small group, all told, because it’s limited to the set of organisms which are both oxygen-tolerant (not something you can take for granted when talking about bacteria) and non-pathogenic.

But the past century has seen revolution after revolution in biotechnology, and—while it’ll still be a long time before we can build a better chicken—we already have all the tools we need to start exploring the rest of the tree of life, to find out what kind of strange and wonderful fruits it bears. What kind of foods could we cook up with a Bacteroides or a Coprococcus as a starter culture? It’s only in the past fifty years or so that such a thing has become possible—but now, the only constraints are our imagination, and our willingness to embrace these new technologies and get creative with them.

The next new fermented food is somewhere out there. And when we find it, it’s gonna make lab-grown meat look like chopped liver.

The Problem With Eukaryotes…

Half the point of lab-grown meat lies in the potential for improvements in “feed efficiency”: how many calories of e.g. soybean you have to put into a system in order to get a tasty calorie of meat out. In a regular chicken, a lot of those calories go into producing non-tasty things, like feathers and blood and eyeballs—so if you can manage to grow just the muscle cells in a big vat, you can theoretically make big improvements.5

Unfortunately, a lot of the non-tasty bits are important for keeping the muscle cells alive.

Some of those bits are easy enough to account for with engineering: Heat the tank with electricity and you don’t need feathers for insulation. Stir it a bit, and you circulate nutrients that would ordinarily be ferried to the cells by blood. But other things aren’t so easily replaced.

Take growth factors, for example: animal cells rely on complex signaling proteins from the rest of the body to know when to grow and divide—the only kind of cells that are always switched to “grow” are cancer cells.6 Those signaling proteins are complicated, which makes them costly to produce artificially…but at least they’re possible with current tech. Other things—like the immune system—are practically irreplaceable. Because the different types of cells in our bodies are highly specialized for given functions, a vat full of muscle cells has no defenses against the germs that an ordinary animal fights off without a problem; working with animal cells in a Petri dish, a single stray cell of Mycoplasma can ruin your entire experiment. At an industrial scale, the problem is even worse: if one tank in a factory turns into a breeding ground for contamination, odds are your entire operation is toast.

We can imagine clever solutions to the problem: maybe there’s a way to keep a few immune cells in the tank along with the muscle cells, and let them patrol for any invaders. That would reduce the risk of a catastrophic failure, but it would also cut into efficiency: those immune cells need food too, and probably some weird growth factors of their own. Take this line of thinking to its logical conclusion—adding back in cell types and structures, sacrificing efficiency to improve reliability and ease of use—and before long you’ll find that you’ve reinvented the whole chicken.

Having hyper-specialized cells like this is the hallmark of eukaryotic life. It’s key to our ability to build complex and wonderful structures like brains and bones. But that hyper-specialization means that the individual cells—and especially things like muscle cells—are passive, fragile, and needy on their own. None of it works without the rest: the muscle cells expect the GI tract to be there, to break down food into individual amino acids; the circulatory system to carry those nutrients to them, fat cells to keep them at the right temperature. Take an individual cell out of that context for even a few minutes, and you have to do a lot of work to keep it alive.

All this makes your average bacterial cell look like Rambo, in comparison. Where you and I have a thousand hyper-specialized co-dependent cell types in our bodies, bacteria are scrappy generalists: every cell is a whole organism, an individual, ready to fend for itself. This is what is meant by “single-celled organism”—and it’s a powerful paradigm.

Rather than relying on an immune system to protect them from invaders, they produce acids, or alcohols, or antibiotics in order to make the environment inhospitable to their competitors—and this is part of why fermentation has been a key part of food preservation for thousands of years.

Think about blue cheese, for christ’s sake! You take this hunk of protein, carbs, and fat, leave it in a dank cave for a few months, and it comes out better than when it went in! “Oh, looks like it made a friend in there!”

Bacteria don’t need any coaxing by special growth factors in order to continue growing and dividing; as long as there’s more food and more room, they’ll keep it up. And if they’re running out of either, half of ‘em can just turn on their flagella and motor off in search of greener pastures. They’re not picky eaters, either—purified amino acids? Pah! Just throw some ground beef into the test tube, they love the stuff.

They’re hardy little guys, too. Just last week, I plated a sample onto a petri dish, and accidentally forgot to put it into the incubator before I left for the night—so it sat at room temperature, when it was supposed to be at 98°F. Animal cells left alone, thirty degrees below their preferred temperature for 14 hours? Forget it—they’re toast, they’re trash. But the next morning, I popped the plate into the incubator, and the day after that, we had growth. This is the beauty of each cell being its own organism, as self-sufficient as possible: they simply don’t know the meaning of “throwing in the towel”.

Of course, for all their virtues, this is still an apples-to-oranges comparison. At the end of the day, no bacterial byproduct is likely to make a convincing meat substitute….but who cares? What if—rather than struggling in vain to make imitation meat—we set out in search of something even better? We should be applying modern technology as creatively and experimentally as we can…rather than fixing our sights on something as boring as “meat, but from a vat”.

In many ways, the effort to make lab-grown meat a reality is an effort to get animal cells to grow and behave like microbial cells would—and at present, this takes an insane amount of effort. At a certain point, you have to ask: Is it worth it, to take over the job of extracting the protein from a soybean, hydrolyzing it into individual amino acids, then purifying and sterilizing those and stirring them into the vat, so that you can grow some chicken cells—when the chicken would do a pretty good job of all that, if you just tossed a few handfuls of beans over in the general direction of the coop?

And yes, if you go to all that trouble you can enjoy a chicken nugget without taking a life, but if we really want to wean people off of animal-based agriculture—whatever your motivation for wanting that—we need to present them with better alternatives. Hell, tofu isn’t even very good, but whoever invented it has probably done more for that cause than PETA ever could.

Fortunately, we have a frankly unbelievable toolbox to work with.

The Nose Knows

Taste and smell speak to us at a biological level. For instance: Check the ingredients on a bag of octopus-flavored potato chips from Taiwan, and you'll see "5'-ribonucleotides". Why? Because RNA is fucking delicious. Your body uses a lot of it, so your taste buds have evolved to be very sensitive to it—and those sensors are wired straight into your reward and satiation centers. The body knows what it needs, which is why the most irresistible foods are usually loaded with calories, salt, and other things that keep us going.

So if we want to find our starter for Cheese 2.0—i.e. microbes that are likely to be both safe to eat and delicious (because they possess beneficial and nutritious chemistry)—we need look no further than the human gut. It might sound gross to suggest using a bacterium isolated from poop to make a new fermented food...but only until you've smelled the result.

I've worked with a lot of different gut bacteria in isolated cultures. Some of them smell as foul as you'd expect from an organism isolated from shit...some even worse. Once in a while, though, I've uncapped a test tube to dump its contents, caught a whiff…and been absolutely entranced.

Is that...pickles? tuna salad? Neither is quite right; there was no fishiness, just a pure and enticing umami aroma with a tangy, fruity undertone. Some blend of short-chain fatty acids along with other, more obscure volatiles. It's never been more than a passing intrusive thought, because there are certain things you can’t do in a BSL2 lab—but my body was very clearly screaming drink it, DRINK IT, that’s the GOOD stuff!, and I have a feeling it was right.7

There are thousands of species in the healthy human gut: a practically infinite number of strains with unique and interesting chemistries out there, waiting to be found. Practically nobody seems to be looking. But a few weeks ago, on a hunch, I inoculated a strain of Coprococcus eutactus into some coffee that I had souped-up with a little nutritional yeast and collagen. After a few days in the incubator, it came out enhanced with an unmistakably buttery note, thanks to a little molecule called acetoin that Coprococcus makes. That’s how easy it can be—the definition of low-hanging fruit!8

Could have been a Trappist

In one sense, calling lab-grown chicken "Franken-meat" is a cheap shot. It's not green, it probably won't kill you, and I’m not holding my breath for a tap routine with Gene Wilder. But Frankenstein undeniably is the story of a man trying to create life: going to unnatural extremes to acquire the power which biology holds for herself—and finding that his best efforts can produce only a monstrosity. It reminds me of the story of a medieval monk, who spent practically his whole life walled up in the monastery tinkering with gears and cogs and springs, and after forty years unveiled his creation: a clockwork monk, a few feet high, that could walk and kneel and offer silent prayers. A marvel for its time…but the first time I heard that story, I thought “Man, you could have just had a kid”.

That’s the foreseeable future with lab-grown meat. Not a monstrosity, and maybe even a technical marvel—but a sad shadow of the real thing. Because if you zoom way, way the fuck in, the real-live monk is a clockwork monk. Collagen coilsprings give your skin its bounce. In your mitochondria, a proton-gradient breeze turns the windmill blades of ATP synthase, clicking a phosphorous into place, charging your batteries. Protein strings fold into sheets, which fold into little origami machines, which produce more protein strings. It’s all physical, just optimized to a scale we can barely fathom—and in our efforts to imitate it, we’re only a little better off than the fumbling monk.

But why try to imitate nature, when we can innovate with it? “Meat 2.0” is ambitious, sure, but so was Dr. Frankenstein—who might have had a lot more luck creating life, if he’d simply set his sights a little lower and found a nice girl to settle down and have kids with. After all, at the end of the day, which monk brought more delight and wonder to the world: the one toiling away on his little toy man, or the lucky Trappist who discovered the miraculous funk of a Belgian tripel?

—🖖🏻💩

Most of this piece was written over a year ago, as part of the Ideas Matter fellowship, sponsored by . Many thanks to Niko McCarty and the whole team at Asimov for their efforts, and many apologies for letting this sit in drafts until the publication was defunct. If you liked this piece, please click the button below to subscribe for free so that future posts come straight to your email inbox.

Subscribe now

If you’re a new recruit, you may not be aware: I don’t just talk shit, so to speak. Microbiome science is my full-time job. My company, Constellation Bio, is focused on turning the kinds of ideas you’ll find in the archives of this blog into actual products—making the other 99% of the human microbiome available as probiotics.

That means we’re always looking for brave volunteers to help us beta-test the future of biology. Some aspects of that (like trying out a new species as a probiotic) are only feasible with participants who are local to our lab here in Berkeley, but anytime we’re spinning up a study that’s amenable to nationwide participation, I’ll put out a call here.

For instance, many of you have already signed up for the ongoing Astrobacillus project, investigating the relationship between serum cholesterol and fecal coprostanol. If you haven’t—and you’re an adult in the US who’s had a lipid panel done in the past year—I’d encourage you to sign up here.

But while that project is underway, we are cooking up some exciting things in parallel, and to that end it’s my pleasure to invite you today to participate in a new study, which we’re calling the Cassiopeia program.

New Micronutrient Just Dropped?

Cassiopeia is a first-in-humans study to assess the safety of a novel dietary supplement containing a mineral which, historically, was thought to play no role in human biology.

But recently, scientists have discovered that a variety of human gut bacteria—including some very important ones like Faecalibacterium—have enzymes that use this element as a cofactor. As I’ve said before, the microbes indigenous to the human gut are almost as much a part of our bodies as the mitochondria: they’re external to the 23 chromosomes, but we have been coevolving with them for so long—and they play such an integral part in our biology—that there’s no sense in drawing distinctions between “our” genes and “theirs”.

So a trace element which is important for a ubiquitous and abundant human gut bacterium like Faecalibacterium may well deserve to be considered an essential nutrient for humans, to the extent that its absence might lead to disease. From the little that we know about this mineral’s role in bacterial biology, a deficiency of it in your microbiome might produce symptoms resembling certain chronic diseases that we don’t currently have good explanations for. Animal toxicity studies have already been done, and it seems remarkably safe—we’ll be using doses about a thousand times lower than the No Observed Adverse Effect Level in rats.

I can’t say much more here without giving it away, because we’re trying something new this time, in that I’m asking folks to sign a confidentiality agreement before disclosing the full hypothesis. It’s not like me to play my cards close to the chest, but if you’re curious about the rationale for this, I’ve outlined it in the text of the CDA.

So: If you’re interested in learning more about this mystery mineral and our plans to investigate it, click the button below—the form takes about a minute to read through and complete, and once you fill it out you should get an email with the rest of the program details.

Cassiopeia Program CDA

See you on the other side!

—🖖🏼💩

The genesis of the hypothesis lay in an Australian study from 2024, which used shotgun metagenomics to look at the gut microbiome in schizophrenia, and found some absolutely huge differences between patients and mentally healthy controls:

The authors concluded that this disrupted state was most likely a result of the patients’ medication, noting that the extent of the changes was fairly well correlated with the dose of antipsychotics they were on. I’m skeptical of this explanation for a couple of reasons, though: For one thing, patients who have more severe symptoms get prescribed higher doses of antipsychotics…so if certain bacteria cause the symptoms, you’d expect to see a correlation between drug dose and the abundance of those bugs. For another, there’s at least one study of “drug-naïve” schizophrenics which reports the same microbiome pattern in patients who had never taken antipsychotics. That study also found that six months of treatment with an antipsychotic (risperidone) decreased levels of these bacteria.1

But the thing that initially made me go “wait a sec” is that I happened to recognize a lot of the names on that list, from my days working at Holobiome—a microbiome science startup back in Boston, where we were investigating the gut-brain axis. Faecalicatena gnavus, the Clostridium_Ms, Absiella innocuum—all of these bacteria were on our radar because they’re known to possess an aromatic amino acid decarboxylase (AADC) enzyme—giving them the ability to turn the amino acids tryptophan and phenylalanine into tryptamine and phenethylamine, respectively.

And pharmacologically, this is an interesting fit for some of the symptoms of schizophrenia: tryptamine is the core structure of most of the classic psychedelics (e.g. psilocybin), and phenethylamine acts a lot like amphetamine in the brain. This jives with what we know about the biochemistry of psychosis, in that—given enough amphetamine—an otherwise perfectly sane person will start rambling about vast and ancient conspiracies, trying to pick nonexistent bugs out from under their skin, plotting to annex Austria etc.

Digging into this possibility, I found support in the supplementals of a meta-analysis of metabolomics/metagenomics studies, which showed that—in multiple independent cohorts from around the world—fecal abundance of many of these schizophrenia-associated bacteria is tightly correlated with the amount of tryptamine and phenethylamine in the bloodstream.

The ones that are elevated in schizophrenia are positively correlated with levels of these psychoactives, and the ones that are depleted in the disease are negatively correlated. Neat!2

But I could tell there was more to it: a few puzzle pieces that had yet to fall into place. I speculated a bit at the end of the original post, about what those might be—but nothing felt solid enough to really stand on.

Now it’s time for the rest of the story.3

The Pharmacodynamics Problem

My main point of dissatisfaction with the original theory was that, although phenethylamine and tryptamine act a lot like known psychoactive drugs in the brain, they only do so for a very short time. The enzyme monoamine oxidase (MAO) breaks them down very quickly, meaning phenethylamine’s half-life in the bloodstream is only about 5 minutes, and tryptamine’s is even shorter.

So putting someone on an in-house phenethylamine/tryptamine slow-drip shouldn’t do much (or at least not for very long), unless the doses were massive—unreasonably high, compared to the amount of tryptophan and phenylalanine in our diets.

My first try at patching the hypothesis invoked N-acyl amides, compounds that have been reported to reduce production of MAO, and which are correlated with some of the same bugs found in schizophrenia. But I didn’t like this, partly because the evidence around N-acyl amide production by gut bacteria isn’t very solid. Those same chemicals are used in plastics manufacturing and in industrial agriculture, so they might owe their presence in blood to something like processed food consumption—or even to the tubes that the blood is collected in.4

So it was a few weeks after I published the second schizo-post, with this pharmacodynamics problem rattling around in my head, that I went back to the supplementals of the Vasileva et al paper and started browsing. One of the tables in their datafile was a metabolic pathway analysis, which I hadn’t looked too closely at the first time—I was too gobsmacked by the differences in species abundance to pay much attention to anything else.

But going beyond taxonomy (i.e. describing things in terms of their genus and species) is often critical in microbiology, because there’s an astounding amount of “give” in the definition of a bacterial species relative to the kind of organisms we’re used to at the macroscopic scale. Thanks in large part to their ability to pick up genes from the environment wholesale and put them to use, two strains of E. coli can have as little of their genome in common as a partridge shares with a pear tree, and this is where it can be helpful to say “forget about what it’s called, tell me what it can do” and directly compare the abundances of individual genes, or the pathways they comprise.

And looking at which gene pathways are differentially abundant in schizophrenics vs. healthy controls reveals a surprising signal—one that finally made the rogue puzzle piece quit its rattling and snap cleanly into place.

“Ethanol degradation”—two different kinds!

Now, you’re probably thinking: “So what? schizophrenics probably drink more.” But I don’t think that’s what’s going on here, for a couple of reasons.

For one thing: they told us so. Self-reported alcohol use was actually lower among schizophrenic patients than healthy controls. And yes, patients lie to doctors all the time, but consider that this study was done in Australia, where I imagine it’s actually quite difficult to keep pace with the “cognitively normal” controls, in terms of alcohol consumption, if you’re playing with the handicap of a serious mental illness. What’s more: alcohol is absorbed very efficiently in the stomach and small intestine, and metabolized in the liver—so while heavy drinking can change the composition of the microbiome, studies have shown that it does so by flooding the system with acetic acid—the end product of alcohol metabolism.5 You wouldn’t expect it to lead to an increase in ethanol degradation genes.

So if there’s any substantial amount of alcohol degradation happening in the microbiome, that means alcohol production must be happening there as well. And while we typically think of alcohol production as a feature of yeasts like Saccharomyces, the capacity is also pretty widespread among bacteria, including some of the bugs you find in the human gut. And in particular:

Now, if you’ve been following along through Parts I and II of this three-part Christmas special, you’re pretty well prepared to understand where this goes next.

See, when alcohol gets broken down for energy—whether by your liver or your gut bacteria—it first gets turned into the toxic acetaldehyde, then into the harmless acetate. It’s in that between-time when pretty much all the unpleasantness associated with drinking, from the carcinogenicity to the hangovers, arises.

This is why people of East Asian descent typically have a harder time with booze; they have a less-efficient version of the enzyme that performs that second step, meaning every molecule of alcohol spends longer wreaking intracellular havoc as acetaldehyde. This is also how disulfiram—one of the most effective drugs to treat alcohol use disorder—works: inhibiting that second step turns even a single drink into a recipe for an instant headache and hangover.

But as we discussed in the last couple of posts, when you have an aldehyde like acetaldehyde and an amine (like tryptamine or phenethylamine) in the same solution, the aldehyde can spontaneously stick to the amine and cause it to cyclize. In the case of tryptamine, the thing that results is a beta-carboline. With phenethylamine, you get an isoquinoline.

And this is amazing because beta-carbolines and isoquinolines are known to act as MAOIs. So think about this in the context of ayahuasca, the South American psychedelic tea (of “one-shotting turbonormies” fame). The active ingredient in it is DMT—dimethyltryptamine—but DMT isn’t orally active on its own, because MAO in the liver destroys the drug before it gets into circulation. That’s why the essential second component of ayahuasca is an herb that contains a natural beta-carboline, to run interference with your liver and give the DMT a chance to get to the brain.

As a result, a bacterium which produces both tryptamine/phenethylamine and alcohol (and thus aldehydes, and thus beta-carbolines) seems like a two-in-one recipe for psychosis, giving you both the psychoactive main ingredients and the adjuvant that makes them long-acting enough to cause problems.

This would also explain why the disease can’t necessarily be pinned to one bug: imagine Blautia producta has the AADC enzyme, but not the ethanol-production pathway, and Streptococcus parasanguinis produces substantial amounts of alcohol, but doesn’t have the AADC enzyme. If you had high Streptococcus alone, maybe you’d get Fatty Liver. If you had high Blautia alone, maybe you’d have something like depression or ADHD, since elevated levels of phenethylamine are thought to cause the brain to produce more MAO—deactivating neurotransmitters like serotonin and dopamine in the process. If you had both, then you’d have a recipe for the Bad Soup.

So this is another manifestation of “the problem with beta-carbolines”. It’s not that nobody ever imagined that they might be involved in psychiatric disease; it’s that—as discussed in the first post of this series—most of the work to investigate this possibility was done before we discovered that they can form spontaneously under conditions routinely used in chemical analysis. And when that unfortunate fact was found out, the neuroscientists collectively breathed a heavy sigh and basically went home, leaving the hypothesis for dead. But has anyone gone back and checked, using the proper methods discussed in that first post, whether schizophrenics have more beta-carbolines in their bloodstream than healthy controls?

The answer, as far as I can tell, is no.

Predictions

Now, it’s already known that schizophrenia—particularly the paranoid subtype—is associated with higher-than-normal levels of phenethylamine, and this aligns nicely with the inferences we can make by combining Vasileva et al’s microbiome data and the metagenomics/metabolomics dataset: if schizophrenics tend to have 45x the normal amount of Ruminococcus gnavus, and fecal Ruminococcus gnavus levels are tightly correlated with bloodstream phenethylamine levels, it follows that schizophrenia would be associated with high bloodstream phenethylamine levels.

This is why I subtitled this post “Some armchair-testable predictions about psychotic disorders”: one of the best ways to explore a new hypothesis is to see if the predictions it makes are consistent with the data that’s already available to us. So what else would this theory imply?

Well, let’s plot this all out, just to keep tabs on the major players here.

We’re proposing that schizophrenic patients’ gut bacteria produce too much phenethylamine and tryptamine, and that they’ve also got some microbiome-derived beta-carbolines/isoquinolines floating around, which prevent MAO from breaking down phenethylamine into phenylacetate. So, in addition to high phenethylamine, we’d expect schizophrenics also to have lower-than-average levels of phenylacetate, right?

And interestingly enough:

Schizophrenia is associated with low phenylacetate—less than half normal! But the paper I took the above clipping from also looked at patients with schizoaffective disorders, and found that they had high urinary levels of phenylacetate. What do we make of this?

Well, “schizoaffective” refers to patients who have symptoms of both mood disorders and schizophrenia. Bipolar is a good example: patients in a manic phase often spiral into a psychosis that’s clinically indistinguishable from that of schizophrenia.

This suggests a model of two diseases: maybe both schizoaffective and schizophrenic patients both have too much phenethylamine production happening in their guts, but only the schizophrenics have a substantial amount of aldehyde—resulting in the MAOIs that prevent these compounds from being broken down.

This naturally implies a heavy role for diet, as well: if you had a lot of a “two-in-one” bug like Enterocloster bolteae or Ruminococcus gnavus in your gut, but ate a strict carnivore diet—meaning no carbohydrates to ferment into alcohol/aldehyde—you might expect it to do wonders for your mental health.

And these observational tests are cute, but still pretty circumstantial. If you want to really put a hypothesis to the test, you’ve gotta get interventional.

So imagine we were feeling evil. Because in this model, psychotic disorders are just the high tail-end of the abundance distribution of a couple of common gut bacteria. That means there’s some subset of the population who only have one or two of the multiple requisite “hits” working against them—or maybe they’ve got all the necessary ones, but not that bad—only 10x average abundance of E. bolteae or R. gnavus. And so they’re struggling with some run-of-the-mill mental health issues, but they’re getting along more-or-less fine. How could we tip them over into not-fine territory?

Well, theoretically anything that increases the amount of acetaldehyde in the gut should encourage the formation of more beta-carbolines, right? And as luck would have it, there are drugs that do just that! We just learned about one earlier: Disulfiram, AKA Antabuse—the pill that takes all the fun out of alcoholism.

So this is our next armchair-testable prediction: If our model were correct, and there’s endogenous alcohol production and degradation happening in the gut, it shouldn’t matter whether patients are drinking: we’d expect psychosis to be an occasional side-effect of disulfiram.

And as it turns out…

This paper from 1967 reports acute psychosis as a side effect of disulfiram in between 2% and 20% of patients taking disulfiram!

Now, this is potentially a point in favor of the model, but it isn’t as clean-cut of evidence as I’d like. It turns out that, because one of disulfiram’s metabolites chelates copper, the drug can also inhibits dopamine beta-hydroxylase, the enzyme that converts dopamine into norepinephrine. Findings around this are somewhat inconsistent, but there’s a world where elevated dopamine levels resulting from DBH inhibition contributes to this phenomenon more than spontaneous MAOI formation. To suss out the real driver, you’d have to look for beta-carbolines and similar compounds in the bloodstream of someone in disulfiram psychosis.

So let’s look a little upstream. The crux of the hypothesis is that bacterial AADC in the gut is responsible for the disease. If that’s the root of where things go wrong, the ultimate testable prediction is that giving someone a drug which inhibits bacterial AADC enzymes should bring a person out of psychosis and keep ‘em that way. Of course, you’d need to find a drug that only hits the gut-bacterial version, because the same kind of enzyme is what lets your brain convert L-DOPA into dopamine and 5-HTP into serotonin.

But as luck would have it, there’s already an FDA-approved drug that works by inhibiting AADC. What’s more, it doesn’t cross the blood-brain barrier—meaning it should interfere with bacterial AADC in the gut, without inhibiting neurotransmitter synthesis in the brain.

It’s called carbidopa.

Carbidopa’s main use is as an add-on in Parkinson’s disease, a neurodegenerative disorder involving a breakdown in the dopamine biosynthesis process. Because the neurons that let you control your muscles run on dopamine, deterioration of that signaling machinery results in the disease’s characteristic tremors. By giving someone L-DOPA, which is converted to dopamine by AADC, you can restore dopamine levels in those motor neurons enough to stave off the symptoms of the disease pretty effectively, at least for a while.

But if you just give someone L-DOPA by itself, a lot of it goes to waste—getting converted into dopamine in the gut and liver before it has a chance to get into the brain where it can help with motor-neuron signaling. And while L-DOPA crosses the blood-brain barrier, dopamine itself doesn’t; it gets scrubbed out by MAO. Hence: giving someone a “peripherally restricted” AADC inhibitor like carbidopa along with their L-DOPA ensures that more of the latter reaches the brain intact.

And this is why, if you bring up “carbidopa for schizophrenia” to anyone who works in neuropsychiatry, they’ll probably balk. At first glance, it’s a counterintuitive idea: carbidopa’s job is to increase dopamine levels in the brain, and schizophrenia is generally understood to be the result of too much dopaminergic activity. After all, the most effective antipsychotics are all dopamine antagonists—shouldn’t carbidopa make things worse?

But if our hypothesis about bacterial AADC is correct—and provided that the AADC enzymes of various gut bacteria are similar enough to the human version to be susceptible to carbidopa’s inhibitory effects—you’d expect carbidopa to hit the disease basically at its root cause. No bacterial AADC, no tryptamine and phenethylamine, no MAOIs. But the problem with “armchair” hypothesis-testing is that it requires someone to have actually tried the thing—and because this sounds, naïvely, like a terrible idea that would make patients worse, I wasn’t particularly optimistic when I went looking in the literature for case reports.

But lo and behold, there is one.

Back in 1984, some research-clinicians at the NIH were lucky enough to have a patient with both a psychotic disorder and a history of stimulant abuse—and they were wise enough to listen to her when, feeling a break coming on, she told them: “Guys, I’ve taken a lot of speed in my day, and whatever’s going on here feels a lot like speed”. Understanding the similarity between amphetamine and phenethylamine, knowing that she had elevated urinary concentrations of phenethylamine, and reasoning that this must come from decarboxylation of phenylalanine by AADC, they tried carbidopa.

It worked. After 7 months free of psychosis, they did a double-blind withdrawal test; she became delusional again, almost immediately—and then recovered almost as soon as the drug was reintroduced.

It would be another thirty years before Fischbach et al reported that some common human gut bacteria possess an AADC enzyme that’s functionally similar to the human version, giving us the theoretical backing to understand why this worked—and why it might work in other patients just as well as it did in her. But knowing what we know now, and especially in light of the fact that the key signature of the schizophrenia microbiome is an overabundance of phenethylamine-producing bacteria, it seems like it deserves a more thorough exploration.

It’s not fully in the wheelhouse of my vision for Constellation, but I’m filing a patent around this one—something like “Use of inhibitors of bacterial aromatic amino acid decarboxylase for the treatment of psychiatric and mood disorders, particularly those with psychotic features”. Finding a new use for an approved drug is a biopharma investor’s wet dream, because 90% of the cost and failure-risk of bringing a new drug to market is in the Phase I and Phase II studies…things that are already taken care of for an existing drug like carbidopa or its Canadian cousin benserazide.6 If we can get a new treatment out there that addresses these disorders in a more fundamental way than existing drugs—and for a fraction of the cost and effort of bringing a new chemical entity to market—I feel like we’ve practically a moral obligation to make that happen.

More importantly, though, if we’re willing to get out of the armchair, this is information that any of us can use now. In the US, doctors have enormous leeway in the form of “off-label” scrips: any approved drug can be prescribed for any condition, as long as the prescriber has some rationale to think it’ll help. They already sell carbidopa as a standalone drug (i.e. without L-DOPA), under the trade name Lodosyn. If you or a loved one are struggling with a psychotic disorder and your psychiatrist tends to be open to experimental approaches, you can bring them the case report—hell, send them this post—and ask if they’ll let you try carbidopa. If they do, drop me a line and let me know how it goes.7

I’ve already devoted a lot of words to this topic, and spending too long with an idea in theory-space without bringing it out into the real world through experiment invariably starts to turn you into this guy—

…so for now I’ll just say thanks for reading—and watch this space. You’ll hear more on this one when I have data.

—🖖🏼💩

It happens all the time: in the course of your usual work, you stumble across some intriguing and unexpected new phenomenon. You investigate, get some interesting results, maybe devise a theory around it, and you’re halfway through writing up what you’re sure is going to be an award-winning publication when you find out that—oh, oops! Turns out this is already a well-known thing in an adjacent field (or in your own field, 50 years ago), and the reason you hadn’t heard of it is that it’s a lot less exciting than you initially thought.

It’s gotten easier to avoid this embarrassment in recent years with the advent of Google Scholar, Pubmed, etc., but the story of the cluster headache that artefactual beta-carbolines and isoquinolines caused in neuroscience isn’t very well-documented online; I only know so much of it thanks to this Ancient Tome my dad gave me a few years back when I was old enough to start asking those questions.

And that’s why I wasn’t surprised to see the same story playing out in the headline of a paper which came out around this time last year, from a new lab at Harvard:

For context: While the intestinal microbiome is as dense and diverse as a jungle, vaginas are a lot less complicated.1 A healthy vaginal microbiome is typically dominated by one or two Lactobacillus species, and much of anything else is seen as a sign of trouble (although there are exceptions to this rule, especially when you start to look at non-Western populations).

Since Lactobacilli are best known for producing lactic acid, it’s widely thought that they serve to acidify the vaginal microenvironment, and thus help keep it free of pathogens: things like Gardnerella and Candida, common culprits in bacterial vaginosis (BV) and yeast infections respectively, have a much harder time growing at low pH. Over the years, studies have found that some vaginal Lactobacilli also produce things like antimicrobial peptides and hydrogen peroxide, which may contribute to this same end.2

This theory does a pretty good job at explaining what we observe when we study the vaginal microbiome in health and disease. For instance: taking antibiotics is associated with subsequent yeast infections, suggesting that stripping away the protective Lactobacilli leaves us susceptible to fungal overgrowth. A Lactobacillus-dominated vaginal microbiome is associated with lower pH and better outcomes in terms of fertility, likelihood of pregnancy complications, and disorders like PCOS. (This is as good a place as any to note that, although that’s the prevailing theory, there’s an alternative model which does about as well at explaining these observations, rooted in the little-known fact that the cells lining the interior surface of the vagina are decorated with proton pumps: the same molecular machinery that produces the strong acid in your stomach. Unlike in the stomach, though, the proton pumps in the vagina turn on primarily in response to estrogen. Since Lactobacilli are among the most acid-tolerant species in any microbiome, there’s a world where Lactobacillus dominance is a consequence, rather than a cause, of a highly acidic vagina. In this model, you’d expect hormonal issues to lie at the root cause of things like recurrent Candida, fertility problems, PCOS etc. with the connection to antibiotic use arising from the gut microbiome’s impact on estrogen metabolism. This is a significantly less optimistic model, in that the problem would be a lot harder to fix.)

In any case, you can see why someone studying vaginal Lactobacilli might be delighted to find that they apparently produce a bunch of interesting chemicals like beta-carbolines. If those chemicals do something consistent with our understanding of the organisms that produce them, like suppressing inflammation? Even better!

But naturally, given the cautionary nature of the tales in the Ancient Tome, I was suspicious from the title alone. By Figure 2, I was pretty sure that the entire paper is bunk.

The thing in the image above that set off the real alarm bells for me is the furfuryl group—the dangly little pentagon guy with an oxygen in it. This is almost identical to an aldehyde that forms when you dry out sugar-rich organic matter. It’s found in things like coffee, oats, and—in all likelihood—the kind of just-add-water dehydrated nutrient broth that they used to grow their Lactobacilli for this study. Also found in that powder? Tryptophan!

Mix them together and boil it down, and you’ve got a recipe for the chemical labeled BC3 up there. Some minor variations on the theme, like using the dipeptide Trp-Val in lieu of tryptophan, or 5-furfural (another compound that can result from cooking certain sugars) for the aldehyde, and it’s easy to see how you get the whole zoo of compounds. Extract 14 liters of broth in ethyl acetate, as reported in their Materials & Methods (yielding acetyl esters BC7&9), evaporate to dryness, and you’ve apparently got a cover story in a Cell-adjacent journal.

Interestingly enough, while some of these chemicals hadn’t been reported before, BC6 already had a name—perlolyrine—because it had been previously identified in a variety of heat-treated foods.

The authors nod vaguely in the direction of this possibility, acknowledging: “Spontaneous production of β-carbolines has been shown to be enriched under acidic conditions. It remains unclear whether lactobacilli production of β-carbolines is driven by an as-yet unidentified enzyme or spontaneous production or both”.

But knowing that they form spontaneously, it’s worth asking: are we sure the bacteria are even involved? Are we maybe just looking at Maillard products, here—like the brown crispy layer that forms on your toast? And in reading through the paper’s methods, I was amazed to discover: they didn’t actually check!

Reader, they did not have a negative control!

At no point did anyone on this team of 20+ Ivy-League students, staff, and faculty think to check whether these chemicals were there in the un-inoculated broth. (Or, if they did, they didn’t report it).

Instead, they tested these nine varieties of basically-caramel by exposing epithelial and immune cells to them in a petri dish. They found that perlolyrine had some anti-inflammatory (or, less charitably, immunosuppressive) effects.

Then they gave some mice genital herpes and smeared the stuff in their vaginas. It sort of helped!

Reader, what the fuck!

in vivo, in vagino

Fortunately, the study’s other headline finding seems to save the day: they actually detected a few of these compounds in the human vagina. Even better, people with bacterial vaginosis (BV) had less of them! That’s practically case-closed then, isn’t it?

Unfortunately, we don’t buy this either. Take a look at the figure where they report this finding:

First off: notice that the concentrations are in the picomolar range. That’s literally a thousand times lower than the lowest concentrations they used in their in vitro and mouse work. It’s a clue that something is off. A further hint: these compounds aren’t even present in most of their subjects, healthy or otherwise.

Fortunately, the authors posted their raw data in the supplementals!

Here, the samples are grouped by Nugent score, but—on a hunch—I sorted them by sample number, re-indexed them from 1, then plotted the reported concentrations.

Dropping the non-detects, the correlation coefficient goes up to ~0.9.

Somehow, beta-carboline concentrations are almost entirely a function of sample number!

This is consistent with a model where amino acids in the vaginal samples react with some aldehyde contaminant in the chromatography column, consuming it in the process to form the detected compounds. If the samples were run more-or-less in number order, you’d expect the highest concentrations of these products in the first few, steadily declining until they’re undetectable in the entire back half of the sample set.3

Funnily enough, this also explains the loose association between beta-carboline levels and BV status. Consider: it’s always easier to find healthy controls than disease subjects for a study; Nugent 0s are a dime a dozen. As a result, the low sample numbers—early recruits—are more likely to be healthy, i.e. have low scores. If beta-carboline concentrations are a function of sample number, and sample number is partially a function of Nugent score, it follows that beta-carboline concentrations would also be a function of Nugent score—although you’d expect it to be a much less robust correlation. And this is what we see.

It’s hard to imagine the authors didn’t notice this sample-number effect. Hard, but not impossible: If they had, you’d think they would make the raw data “available upon request” rather than just uploading it for all the world to see.

But charitable interpretations only go so far. Look at the line I added to the first graph in this section—the LLOQ. That expressly indicates: “below this line, we don’t trust the numbers”. Even so, those numbers get used in their analysis, to produce that two-star 0.007 p-value. Nobody does that without knowing, in their hearts, that they are doing something wrong. This shouldn’t have made it past peer review, but what else is new?

So I emailed the lead author, inquiring politely at first. Had they actually measured beta-carboline concentrations in the supernatants used in their animal work?

They had! But the results looked like the vaginal samples—picomolar, a thousand times lower than the concentrations necessary to produce an effect in vitro.

So they had this data, they had run the experiment, but they didn’t include it in the paper because it contradicted the narrative they were building. After a little more correspondence, I shared my biggest concerns, asking—somewhat apologetically—if they had noticed the sample-number effect, and explaining that something similar had happened to us back at Holobiome, when we were looking at N-acyl amides.

She replied that the samples weren’t run in numerical order, which may be true to some extent, but doesn’t change the fact that beta-carboline concentrations are 1000x more tightly correlated to sample number than to the biological variable they were studying. That demands an explanation.

I asked if they had checked the un-inoculated media for beta-carbolines. I got an out-of-office autoresponder. Fine, okay, I am annoying.

And if you’re a newly minted Harvard lab chief, who’s no doubt fought tooth and nail for this place at the top of the prestige heap, the thought of having to retract the first paper your lab ever puts out because it was wrong in some very obvious ways probably makes you want to curl up in a ball and scream.

I didn’t want to do that to her. But I wrote to the journal, asking if they’d publish a response or a reanalysis if I submitted one, maybe a letter to the editor kind of thing. I long for the days when scientists would beef openly in the pages of scholarly journals—arguing and rebutting one another in new open letters with each subsequent issue, letting the readers follow along and decide who to believe. Unfortunately, those uncivil days are gone; Cell Host & Microbe had no interest in publishing such a thing. Understandable! If something really shoddy can slip by your review process, you must not be a very high-status publication. Best to keep up appearances.

I wrote the response anyway because all this went down while we were bumming around Amsterdam on our honeymoon, meaning to send it to another journal, but never got around to it. Maybe I’ll put it up on BioRxiv…but the final section of it follows:

This kind of “fuzzy fraud”—like the selective omission of certain results which contradict an otherwise-clean narrative—is all too common. It is at once understandable, given the incentive structures in modern academia…and unforgivable, being a violation of scientific principles at the most fundamental level. When one begins, in the words of Sir Arthur Conan Doyle, “to twist facts to suit theories, instead of theories to suit facts”, one has become an enemy of the truth.

We are beginning to see a reckoning with the problem of fraud in biology, as evidenced most recently by the demolition of Eliezer Masliah’s body of work and reputation. However, the Masliahs of the world, the unabashed fabricators and photoshoppers—their work represents the mere tip of the Iceberg of Bad Science. The subsurface bulk of it is composed of more-or-less honest mistakes by well-intentioned people, glued together by financial pressure and willful ignorance.

There is a pithy saying, fashioned after Occam’s Razor, which goes something like:

“Never attribute to malice that which can be explained by mere stupidity.”

While charitable in a sense, it doesn’t prove very useful in the face of malice, or even in the gray land of plausible deniability. The altogether less pithy “Hubbard’s Corollary” goes:

“Never attribute to malice or stupidity that which can be explained by moderately rational individuals following incentives in a complex system.”

And this is the crux of it—but as banal and impersonal as this sounds, it nevertheless represents a moral crisis at a personal level. Biology is the frontier of medicine, and in that capacity biologists are entrusted with vast sums of public funds, with the expectation that we’re doing the best we can to cure everything from cancer to bacterial vaginosis. In that sense, even supposedly minor “sins”—like dropping an iffy outlier that you maybe shouldn’t have, or using the statistical test that gives the best p-value, even if it’s not the most appropriate—are ultimately a choice to prioritize advancement and prestige for yourself, borrowing against the suffering of people who are counting on you to help them.

Honoring that trust, as a scientist, is not served by tucking away the loose threads of a narrative, but by pulling and testing each thread in turn, in case it makes the whole thing unravel. It’s a dismaying feeling when it does, but it’s a step closer to the truth, and it’s an invitation to make the next version less like fabric and more like chain mail: crafted one careful loop—one experiment—at a time, each ring interlocking with all those around it such that, should one link fail the test of time, it does not break the pattern of the whole.

And although it is cumbersome and difficult to make, such solid armor has this advantage over cloth:

You will never find yourself suddenly naked in the street, just because some little goblin saw a loose thread and decided to pull.

—🖖🏼💩

Stay tuned for Part III later this week, in which we’ll bring it back around to the schizophrenia hypothesis.

See, the problem with beta-carbolines is that it’s very hard to look for them without accidentally producing some. This means that, if you’re analyzing something squishy and mysterious to find out what it’s made of, it’s hard to say whether there were beta-carbolines in the thing to begin with, or if they only cropped up because you went looking.

And this is a pity, because it’s one of the most interesting classes of chemical compound in biology. Many of them are remarkably potent drugs, capable of doing strange things to your most important neurotransmitter systems—and they DO occur in the wild…but the ease with which they can also appear basically out of thin air means that the research around them is fraught with red herrings and false trails that ultimately lead nowhere.

Even so, the tale of the discovery, then debunking, then re-discovery of endogenous beta-carbolines (and the related mouthful: tetrahydroisoquinolines) is a great story, and an object lesson in the pitfalls of modern science. What’s more: if we can dodge those pitfalls for ourselves, under this pile of what appears to be herring we might find the key to solving some of the greatest mysteries in psychiatric medicine.

Growing Pains

The story starts in the ~’70s, with the widespread adoption of High-Performance Liquid Chromatography, or HPLC—a finicky but very sensitive technique for analyzing the chemical makeup of complex substances by separating them out into their component parts.

As HPLC came into broad use among neuroscientists, people suddenly started finding all sorts of new compounds in the mammalian brain, each begging its own spectacular question.

One was salsolinol, a mysterious derivative of dopamine, which would later be found at elevated concentrations in the urine of alcoholics. Before long, people were speculating that this might somehow be involved in the biology of addiction. The closely related N-methylsalsolinol has neurotoxic effects, and can produce a syndrome that looks a lot like Parkinson’s disease in animals. Could this have something to do with neurodegenerative disorders?

Another was 5-hydroxytryptoline, a derivative of serotonin which acts as a monoamine oxidase inhibitor, or MAOI—meaning it blocks the enzyme that breaks down serotonin and dopamine, resulting in more of these neurotransmitters active in the brain. Some of the earliest (and—if we’re being real—most effective) antidepressants were MAOIs, so people naturally began to wonder: Could this apparently endogenous molecule somehow be involved in the biology of depression?

Still another, known as beta-CCE, turned out to be an inverse agonist at the benzodiazepine receptor, which is the target of anti-anxiety drugs like klonopin or xanax. Longtime readers might remember that, in my “Bacterial Xanax” piece, the benzodiazepine antagonist flumazenil plays a pivotal role in the story—as does the difference between an antagonist and an inverse agonist.

If a person has overdosed on something like xanax, flumazenil can save their life, the same way that narcan can bring someone back from a deadly opiate overdose. But flumazenil is not anti-xanax, only un-xanax: Giving it to a healthy person won’t make them anxious, the same way that giving narcan to someone healthy doesn’t cause them pain.

Beta-CCE is anti-xanax, effectively the real-life version of the fear-gas the Scarecrow uses in Batman Begins. I’ve heard a firsthand account of one of the early tests of beta-CCE in a non-human primate; a rhesus macaque injected with the stuff thrashed so hard in panic that it broke the chair it was strapped to. So naturally, people asked: Could this chemical have something to do with aggression or anxiety?

It was an exciting time in the field; it seemed like a whole new frontier in neuroscience had suddenly cracked open—but as data began to roll in from various labs investigating these compounds, it became clear that something wasn’t quite right. For one thing, there was wild variability in the results, with some researchers finding next-to-none of these compounds in the brain, while others reported them in such abundance that they should have been detectable even without the sophisticated new methods of HPLC. Overshadowing all these discrepancies, the larger mystery remained: How are we just now discovering all these important brain chemicals? Why didn’t we notice them before?

With the benefit of hindsight (and the blood, sweat, and tears of a lot of analytical chemists) it’s obvious: for the most part, we didn’t notice them because they hadn’t been there.

See, HPLC essentially lets you take a complex mixture and separate it into its constituent molecules, based on their size and structure. Done right, you can start with a complex substance like cheese (which might contain a few thousand different chemicals) and end up with all the water from it in one flask, the lactose in another, the cholesterol in a third, and so on. Practically the only limit on resolution is your patience, and whether you care enough to measure what percentage of the triglycerides in your cheddar have palmitoyl- vs. stearoyl-groups at the 2-oxygen.

In practice, this involves dissolving your substance of interest into weird chemicals like acetonitrile, and then pumping it through little tubes full of silica at some insane pressure like 7,000 PSI. On the other end of this ordeal, the sorted molecules are typically then shot out of an ion cannon and smashed to bits, so that scientists can look at the pieces and figure out what they were made of.

The Problem

Here is where the fun begins: The solvents used in HPLC often contain trace amounts of chemicals like formaldehyde, even when they’re advertised as being ultra-pure. Aldehydes are pretty nasty molecules, because they’re very reactive; they stick to things, and alter their makeup in doing so. If you’ve ever been to an open-casket funeral, you’ve seen formaldehyde’s handiwork—it’s used to embalm corpses because it basically glues everything together at a molecular level. This preserves flesh’s large-scale structure, while messing it up enough in a biological sense that nothing can use it as food. If you’ve ever had a hangover, then you have firsthand experience with aldehydes—specifically acetaldehyde. You might be aware: Alcohol itself isn’t that bad for you, and neither is its eventual transformation product, acetate.1 But the halfway mark between those two not-very-toxic compounds is the aldehyde intermediate, and a lot of the unpleasantness of a hangover is a direct result of the compound’s reactivity, its tendency to stick to other molecules, basically at random, and gum up the works of your cellular machinery.

All this to say:

When you put these reactive chemicals into a tube with some dopamine or serotonin, the aldehydes can stick to the neurotransmitters, do a little quantum ballet number, and rearrange like so:

…to form a beta-carboline (or, in the case of dopamine or adrenaline, an isoquinoline like salsolinol).

Now, the amount of aldehyde in your average jug of chromatography solvent is basically negligible. But with the development of the exquisitely sensitive HPLC, a few bad batches of solvent was all it took to send the field off on a series of wild goose chases: anytime people ran serotonin through an HPLC rig using insufficiently pure chemicals, they’d find some amount of its beta-carboline derivative, 5-hydroxytryptoline, even if there was none in the brain tissue their sample started out as part of.

Eventually, people figured out ways to deal with this. Somewhat ironically, one of the most effective is to start by deliberately producing very reaction you're trying to avoid: dump a bunch of serotonin into your solvent first, let it react with any aldehydes to form beta-carbolines, and then filter those out along with the remaining serotonin to obtain a liquid that’s truly aldehyde-free, which you can then use for your extractions.

And if you do it this way, you find that most of the signals vanish. In many of the cases where they still show up, like in the urine of alcoholics, it turns out that their presence can be traced back to dietary exposure: the beta-carbolines in drinkers’ urine mostly come from beer and wine, where they form during e.g. the roasting of grains, the cooking of wort, and the fermentation into booze.

Whatever interesting properties these compounds might have, it seems they’re not made in the mammalian brain in any significant quantity, and they’re not part of its normal workings. As this came to light, research into them continued, but with considerably less enthusiasm.

Still, it’s worth remembering that genuine beta-carbolines DO exist in the wild, though—like in the plant Peganum harmala, which gives the compounds harmine and harmaline their names. These “harmala alkaloids” are the essential second component in the psychedelic drink ayahuasca; their MAO-inhibitory activity runs interference with your liver, letting the drink’s active component—dimethyltryptamine or DMT—slip into circulation and do its thing. We’ll talk more about this in an upcoming post…but first, we’ve gotta talk about vaginas.

Stay tuned for Part II, later this week.

—🖖🏼💩

I swear to god, the next time someone sends me a link to a ChatGPT “analysis” of one of my articles, there’s going to be an unfortunate mixup in my postal logistics-handling spreadsheets, such that all the fecal samples which are supposed to come to my lab that week will instead wind up in that person’s mailbox by mistake.

(Kidding, obviously; I would never let good poop go to waste like that.)

Jokes aside, it’s worth talking about why—even though it’s just one of the many things current “AI” models are bad at—using them to try and understand the human gut microbiome is a uniquely terrible idea. And, at the risk of making a habit out of putting people on blast for relatively minor offenses, a recent example provides a good jumping-off point for the discussion.

A while after I put up the ALS hypothesis post, a reader reached out explaining that the disease runs in his family, although it appears to have spared him so far. So we sequenced his gut microbiome, and did the same for his relatives with ALS. I’m still combing through the data, but I sent him a copy of the big ol’ table reporting the relative abundances of all the microbes in his and his family’s guts, along with the same data from the participants in my ongoing cholesterol/coprostanol study for comparison (They’re not all exactly “healthy”, but to my knowledge none of them have ALS).

I don’t typically do this, because it’s a lot of information that most people can’t make heads or tails of, but he asked with an understandable sense of urgency, so I did some quick number-crunching and sent him my preliminary analysis along with the spreadsheet. Sure enough though, a day later I’ve got the dreaded link in my inbox, prefaced with the usual “Not sure if this is helpful, but I asked [LLM] and…”

Pessimistic, I clicked—like any good scientist should—hoping to be proven wrong. I was disappointed, but with the consolation prize that its attempt was so godawful that it broke me out of a month-long writer’s block.

The reader gave it good context around his family’s story, and fed in the URL of my ALS post along with the relative abundance spreadsheet. After a few overly florid lines of sympathetic therapy-speak and the requisite [I AM ARTIFICIAL INTELLIGENCE, NOT A DOCTOR] disclaimer, it launches in with:

“The most distinct marker mentioned in Skolnick’s article (and supported by several studies, including Herzog et al) is that ALS patients often exhibit a “pro-inflammatory” gut microbiome characterized by a lack of Prevotella and an overabundance of Bacteroides.”

If you’ve read the ALS post, you might notice that this is what AI folks would call a “hallucination”, i.e. it’s dead fucking wrong from start to finish.1 Now, it’s the kind of thing you could half get away with if you were talking about practically any other piece of writing about the microbiome (or neuroscience, autoimmunity, metabolism, or any other aspect of health/disease for that matter), because “characterized by a pro-inflammatory state” is the most generally applicable catch-all excuse for a mechanistic insight imaginable. It could be said of practically any disease state, and—being applicable to everything—means next to nothing.

This is a pretty solid way for the LLM to fulfill the core directive sculpted into its architecture by a zillion hours of human feedback, which can be summarized as “string together words related to the query, in a sensible order, without saying anything logically inconsistent or objectively wrong”. Statistically, it took its best shot based on its training dataset—presumably missing the fact that it was supposed to read the linked article first and incorporate that into the “context window” that it uses to decide what to say.

And this is a great example of why the fundamental structure of an LLM is inherently ill-suited to synthesizing this kind of information.

Say there are six studies out there reporting elevated abundance of Bacteroides in COVID, and another three studies reporting reduced abundance of Bacteroides in COVID. If you ask an LLM about the microbiome’s role in COVID, more often than not it’ll tell you something like: “There’s substantial evidence indicating that Bacteroides is [elevated/depleted] in COVID,” picking one or the other somewhat randomly, because both of those are clearly acceptable words to slot into that spot in that sentence. It’ll pick “elevated” twice as often as “depleted”, because weighting your selections based on how often each one appears in sentences similar to the one you’re constructing is a pretty good way to produce sensible-sounding text; this context-dependent weighting is the basic essence of how all LLMs work.

What this means, though, is that it is not very likely to tell you “There’s conflicting evidence around the microbiome’s role in COVID, with some studies reporting an increase in Bacteroides, while others report a decrease”. Even if someone’s written a review article which says that, you’re not likely to get it as an answer from the LLM, because it looks like a minority viewpoint, relative to the nine individual articles it’s summarizing.

So it will come up with a sentence like “There’s substantial evidence indicating that Bacteroides is [elevated/depleted] in COVID, suggesting that this genus may [exacerbate/protect against] the condition”, because a thousand sentences of that general structure were in its training dataset. By incorporating a few words of context into the frequency weights, you can imagine how it would reliably pick between “exacerbate” and “protect against”, based on what it had already said, to construct something internally consistent: In the average sentence like the one above, “elevated” is almost never paired with “protect against”.

But this example makes it clear that there is nobody home, no understanding, and the appearance of understanding—of “object permanence”—is basically an elaborate parlor trick. The people who wrote the words which went into its training dataset, they had minds—and the shadows of these are encoded in their words. But in any kind of knowledge-work (and in science in particular) the precise meaning of a word and the nuances of the thing it refers to become critically important, such that any random variance you introduce to a text—i.e. the only thing that makes these models different from a search engine, which just displays the original text verbatim—makes it less useful.

The Importance of Discernment

What do you get when you take the average of five wrong answers and one right one?

This is the problem in a nutshell, because—while there’s an extraordinary amount of data out there on the microbiome in practically every disease—most of it is not very good. When you’ve got a situation like our Bacteriodes/COVID example, how do you know whether or not to trust a given study?

Well, were their sequencing methods based on shotgun metagenomics, or 16S? If the latter, right off the bat you’ve lost everything about eukaryotes like Candida, so if those play a role in your disease, there goes your signal. Still, there might be useful information in there, depending on what primers they used. If it was the standard out-of-the-box 27F that you get in a metagenomics kit like Oxford’s, you can basically throw the study in the trash; at least half your Bacteroides species aren’t even going to show up, along with most of the Actinomycetota (which contains good guys like Bifidobacterium and bad guys like Eggerthella).

If they DID use shotgun metagenomics, hey great—but before you trust the numbers, check their lysis and extraction protocols; did they validate with a mock community standard of known composition?

For a concrete example, have a look at the differences in community composition you get when you run the same sample through three different sequencing services:

Now, I had to do some major “flattening” (i.e. cutting out everything below 1% relative abundance) to even make this legible, but it demonstrates the above points nicely. That pink bar, Blastocystis, doesn’t show up in 16S, because it’s a eukaryote. Odoribacter laneus, which 16S says is 21% of this person’s microbiome, is actually closer to 1% by shotgun.

These are the kind of obstacles that stand in the way of our understanding these diseases, which is why LLMs—and even things like protein-folding models built on the same architecture—stand next to no chance of helping us overcome them.

The Cave

My wife and I went to the Bay Area “secular solstice” celebration a couple of weeks ago. It’s basically what you’d expect from the name: a few hundred friendly nerds getting together in a theater auditorium for something very much like a Christmas mass—readings, homilies, songs that everyone sings along to—except it’s things like “The Times They Are a-Changin’” instead of “O’ Little Town of Bethlehem”.

And it always feels good to get together and sing with a bunch of like-minded people, but I particularly love what it represents: the modern atheist movement growing out of its singular focus on intelligence, and into an appreciation for wisdom.

It’s a recognition that, while shaking off the chains of religion is all well and good (to the extent that faith has been used by the powerful to manipulate people for most of human history), the social glue it provides is an important part of human culture, and we’re poorer for its absence. That it’s worthwhile to get together to experience fellowship and community and awe, even if the supreme creator of the universe doesn’t command it.

So I was more than a little disconcerted by the extent to which the organizer’s reflections—the sermon, for lack of a better word—were focused on the existential risk posed by the supposedly impending emergence of superintelligent AI. The general message was that we’re living through a moment in human history much like the Cuban missile crisis, where this new technology threatens to wipe out our species unless we can come together and find a way to stop it.

The general outlook is grim, as best summarized in the title of a recent book speculating on the topic: “If Anyone Builds It, Everyone Dies”. The author, founder of the Machine Intelligence Research Institute Eliezer Yudkowski, was in attendance, and gave a brief speech about staying sane in spite of the end of the world. His words are clearly taken pretty seriously here. At one point, with the stage in full darkness to symbolize the longest night of the year, the organizer expressed the extent to which this fear had gripped him.

“Guys,” he said—voice breaking in a sob, “I don’t think we’re gonna make it.”

And although it’s largely taken as a fact among this crowd that artificial general intelligence (or AGI) is an imminent reality, none of the people who I spoke to afterward could make a convincing case as to why this should be so. Effectively the only evidence offered was the rapid progress made in the development of LLMs in recent years.

This IS impressive, and I get why it’s jarring: practically out of nowhere, here’s this thing that can carry on a conversation as if it were a person. How long before the computer is smarter than us, smart enough to recursively improve itself—after which it can do the kind of runaway-to-infinity that Isaac Asimov imagined?

But this argument from extrapolation doesn’t hold, for the simple reason that language and thought are not the same thing.

Language, in a very real way, is the shadow of thought.

See, all the information that gives an LLM its structure is encoded in the text of its training dataset.2 The rhythms of sentences, the relationships among words, the contexts in which they are used: these features encode information about the shape of the mind that produced a given sentence, just as a shadow tells you something about the shape of the thing that casts it…albeit not very much. Turn your head in profile, though—write another sentence—and we can discern a little more of your shape.

Study enough of these flat projections, and it becomes possible to extrapolate back and make a sculpture of the whole form: a model that casts all the same shadows as a man, indistinguishable from any angle of illumination. With enough sentences, why can’t we do the same thing to create a model of a mind?

And we can! But the thing you get by back-tracing silhouettes is not a man. It’s a mannequin. Even its superficial resemblance to a man is inherently limited, no matter how skillful the sculptor: Where is its navel? Where is your shadow’s?

And while you can fix that particular problem by having someone who knows a thing or two about humans look at it—this is the role of “Reinforcement Learning by Human Feedback”, or RLHF—it’s only the outward sign of the thing’s lack of interiority.

Topologically, you can define concavity and interiority as the properties of features which are necessarily lost in the projection of a form from a higher-dimensional space onto a lower-dimensional surface. And while that sounds complicated, a bowl is a good example of concavity: there is no angle at which you can hold it such that you can judge the depth of its hollow from the shadow it casts. For all you know from its shadow alone, it might be solid all the way through, holding no water at all.

Now, a shadow is cast on a two-dimensional surface, annihilating information about a thing’s depth. Language, being a surface that has almost as many dimensions as there are words in its vocabulary, can capture and represent a great deal of structure. But how many dimensions does a mind have? It must be more; you can tell by the way we have to hold a thought up to the light, talking it out, turning it this way and that, just to describe its outline.

And not only is the mind n-dimensional, it is highly concave. Just look at all those sulci! Secret hollows that leave no trace in its silhouette; places even we only notice when some jagged thing unexpectedly tumbles out of them.

And this is why a sculpture, no matter how long you work its surface—no matter how closely you make its shadow resemble that of the real thing—will not spring to life. Because the capacity to breathe or sing or swing a hammer, anything you might want a man to do, is undergirded by immensely complex structure which has nothing to do with what is visible from the surface, except in that it imposes the constraints that give the surface its shape.

When constructing an argument in text, the logical structure of the thing you want to communicate forms first in the brain’s own wordless language, and only by projecting it out onto our vocabulary does its form become visible to others. A logical argument, the taste of salt, the feeling of sunlight on your skin, l’appel du vide (that intrusive thought about throwing yourself off the bridge as you look down over its edge)…these things do not come to us in words. Practically nothing does. And so there is no hope of replicating, from words alone, the neural architecture that produces them.

This is what Plato was talking about, with his allegory of the cave: the vastness of the gulf between our representation of a thing and the thing itself. The perils of conflating the two—of deciding two things are the same because they cast the same shadows.

So, lacking lungs, the statue will not sing. Lacking muscles, it cannot dance. Some will argue that it doesn’t need to be able to; that AI is going to be a different kind of mind, and doesn’t need to replicate exactly the form of the human in order to be dangerous, or terribly powerful. This much is true. The sculpture does not need to be able to breathe.

To be dangerous, at least in the way that has the rationalists sobbing with despair onstage at Christmastime…it needs to be able to sculpt. And it must be a greater artist than the one that created it, able in turn to create an artist greater than itself.

In some sense, an LLM can sculpt already, to the extent that it is sculpted of software and not of stone, and writing software is a process of generating text. But generating code that is syntactically correct, compiles, runs etc. is equivalent to holding the chisel and moving the hammer. It is not the hard part of sculpting. The hard part of sculpting is seeing the thing within the marble, devising how to bring it out—just as the hard part of making software is conceiving the logical architecture that will achieve your desired result. Maybe it’s telling that the word—conceive—refers to a fundamentally organic process. The only way we know of by which a man can create a greater sculptor than himself starts with another human mind, young and plastic and endowed with intuition.

Statuary and LLMs are alike in that they are both inherently deceptive, whether or not they’re made with the intent to deceive. An LLM’s core design constraint is to resemble a human as closely as possible in exactly one way—text generation—despite the total absence of the interior that makes the surface possible on an organic being.

Take Michaelangelo’s David.

It’s a masterpiece of imitation; you can see the tendons in his hands, the veins standing out under the taut skin of his arm, the anxiety in his eyes as he stares at the approaching Goliath. Oh wait, no you can’t, because all that subsurface structure is just a masterfully executed illusion.

And all this does a great job at short-circuiting our little monkey brains, which are equipped for few things better than guy-recognizing.

So even though it is tempting to make a commonsense appeal along the lines of “Just talk to it! You can tell from how good it is at talking that it’s smart and getting smarter”...this is throwing yourself directly into the trap set by its appearance, making intuitive guesses about its overall complexity based on superficial complexity. This is a trap because your intuition was trained for a hundred million years on natural forms, where a bulging vein has always meant blood flowing beneath the surface, and the ability to carry on a conversation has always meant a reasoning mind.

But just think: how long did it take, after the creation of the first convincing statue, to make one walk? I remember seeing the first video of one of those humanoid Boston Dynamics robots sometime around 2012. And I think it’s telling that, although it has the general shape, the Boston Dynamics robot doesn’t look much like a man at all; the David was not a technical milestone on the road to this artificial man the way that gyroscopes, hydraulics, and electricity were. So while we may be making progress toward the development of AGI, making “progress in LLMs” your proxy for how that’s going more broadly is a great way to guarantee that you’ll overestimate how far along we are.

And knowing that the statue is designed to lead your intuition astray, when someone tells you that the next statue will be able not only to walk but to sculpt statues of its own, even more remarkable than itself, you should not accept this as true unless he can explain to you how they’re going to do the eyes, the muscles, etc. Because while LLMs are technical marvels, they’re also like statuary in that their capacity to perform human tasks does not scale with their superficial resemblance to a human, except in one domain: convincing the viewer that there is a person there.

The Temple

And this too is telling, because human life really did change with the development of lifelike statues, didn’t it? Because now you go to the city and there’s a temple there, with a man inside made of stone. And he’s massive—thirty feet high! You’ve never seen anything like it before, and the guy-recognizer in your brain is going fucking wild. And even though you know, intellectually, that he’s made of stone and he can’t hurt you, you feel a shiver down your spine. You feel awe.

But there’s a man in the temple, and if you ask him whether the statue can walk, he’ll tell you no, of course this is just a likeness—but the real guy is even bigger—unimaginably big—and he doesn’t need to walk, because he is the wind.

And if you point out, quite rightly, that this doesn’t make any goddamn sense—how can the wind be a guy?—he’ll smile knowingly and say Yes, this is the Great Mystery, and that even he does not fully understand it; that there are some things in this world that man is not capable of understanding.

And it sounds innocuous, maybe even wise—but if you buy this, then he fucking has you. You have surrendered trust in yourself, and before you know it you’re bringing your best goat to the temple so the big stone guy who is also the wind can have lunch, and you’re bowing to him so he knows that you respect him, and you’re asking him for help with your problems, and sending your daughter to live as one of his temple virgins. And maybe this last bit makes you particularly uneasy for obvious reasons, but the priest assures you that everything is aboveboard—I mean, c’mon, “virgin” is the whole job title! Obviously nothing creepy going on.

So this is what kills me, because it’s the exact same story, only this time instead of the visual-processing guy-recognizer, it’s the linguistic-processing one. Never before have we had to take executive control of that circuit and suppress it, the way we had to learn to do with images. And we’re getting better at that, but we’re still not even very good—you can draw a smiley face on a rock and you’ll immediately start having stupid little thoughts like “He looks like his name is ‘Clarence’”.

And this instinct is going to be an even harder one to beat with LLMs, because the ability to communicate through language has been the defining feature of our species for the past hundred thousand years! Is it any wonder that a language model basically oneshots people’s psychological defenses against this pareidolia of the mind?

But if you get past the gut reaction and really look…where is the evidence that this path leads toward a mind more capable than our own? To the kind of recursively self-improving machine intelligence that would make spending billions of dollars on chips, or millions on “AI Safety Research” a good use of anyone’s money?

The statue in the temple will never walk. That’s not its job.

Its job is to flummox you for a moment, put you on your back foot. To give the priest his opening to tell you that the tingling on the back of your neck means something more than it does—that it’s a sign of the power in the statue and the thing it represents. To give him power over you. Because at the center of every cult throughout history is someone who’s discovered that some social engineering and a little sleight of hand—maybe a psychoactive plant, or knowledge of when the next eclipse will happen—is all it takes to get nine wives and the best hut in the village.

So it is a little suspicious to me that, such a short while after God was pronounced dead, this new character has pulled up wearing a cardboard box on his head, going “Beep boop, I’m the machine superintelligence, hop in and let’s all go back to my place for some eternal life and the solutions to all your problems! Oh, and don’t forget to give my prophet a bunch of money, otherwise the evil version of me is definitely going to show up and bring about the apocalypse.”

Does this sound familiar to anyone else?

And a lot of rationalists are open about this part; more than once I’ve heard the quest for AGI described as trying to “build God”.

And I get it, because belief in a God fulfills a lot of core human needs.

It puts us at the center of the universe, letting us pretend we’re not small and insignificant in a vast and indifferent cosmos. The stakes are practically infinite! If the machine superintelligence isn’t aligned with human ethics, it will wipe us all out!

It lets us avoid confronting death; the fact that in the blink of an eye our lives will end, as will those of everyone we love—that by the time the galaxy has made a single rotation there will be nobody left to remember us. I don’t want to die, so freeze my body until the singularity comes and we can upload my brain to the mainframe, and live forever among the stars in cyberspace.3

And it lets us imagine ourselves part of a grand plan to repair the evils of the world; lets us tell ourselves our work is meaningful in spite of our ignorance and helplessness. I see the world around me full of suffering, and I don’t know how to stop it—but what if I can create something that does?

But at the core of rationality is the understanding that wanting something to be true has no bearing on whether or not it is. And in spite of the technologists’ earnest belief that they can make an actual god, a sober look at the fruits of their labor thus far makes it clear that this is just another statue of one. The fact that it can talk back to you is a genuinely marvelous feat of engineering, but when you consider that this is a product of a trillion of words of human-written text and a million man-hours of RLHF—fine-tuning it by telling it “No, a person wouldn’t say that; say this instead”—it becomes clear that the underlying principle is just Chatty Cathy meets the Mechanical Turk.4 And unless one of the people working in OpenAI’s RLHF division can tell it the correct answer to give when someone asks how to make an artificial general intelligence, that knowledge isn’t going to spontaneously appear in its neural net.

Look, the digital world is one of pure representation, where the shadow of a thing is all there ever is to see of it. In that sense, it’s the closest thing imaginable to Plato’s cave made manifest. Living in it as heavily as we do, it is easy to forget that the shadow is not the thing. That language is not thought. And it’s particularly tough with LLMs, because for all their complexity they are stuck in the cave; they can’t be taken out of it and examined in the light of day.

So awaken, brothers and sisters. Shake off your chains and leave the cave; you should not cower before this idol which stands in the place where God once stood—and neither should you hope for salvation from it. Admire it for its fineness of form; use it, if you can find a real use for it, but do not trust your eyes in the darkness of this place, where they may be fooled by its likeness to a man. Its hallucinations, as the absence of a navel, show it for what it is: empty inside, a mouth without a mind, babbling echoes. Do not trust the shadow-plays of its priests, who enjoy the harems they would not have in the light of day, when they tell you its glory is soon to emerge from the shroud of a mystery beyond our capacity to understand. These are the same shackles your forefathers died in, waiting here in the flickering dark.

Trust your reason, your mind. There is nothing in the world like it.

—🖖🏼💩

In these conversations, I’ve been clear on what I see as one of the most exciting things about what I’m building: the potential to have long-term impacts on people’s health with a short-course intervention. If we can get engraftment—where an introduced species really takes root in your gut and joins your microbiome indefinitely—it’s a whole new ballgame.

See, if introducing a cholesterol-reducer works as well in humans as it does in animal studies, it almost doesn’t feel right to call the product a “probiotic”. It’d still meet the technical definition of the term, but it feels like a fundamentally different category of thing than the freeze-dried yogurt-starters-in-a-pill that you take once a day (and which, by and large, do nothing). If we’re successful, the end result has less in common with Culturelle than it does with LASIK, or orthodontia: semi-permanent upgrades for your body. I’m chewing on what to call this category (I think “Trait” captures it nicely), and if you have suggestions feel free to leave them in the comments, but that’s not really the point of this post.

See, I’ll never forget the first time I got this one piece of feedback from an investor. I’m on a video call with this potential angel; an English guy, very blue-blood vibes. I explain the vision to him, and at a certain point in the conversation he sort of steeples his fingers and says in his posh accent:

And I’m glad that, the first time someone said something like that to me, it was phrased in such a way—and said in such an accent—that it was downright comical, how obviously evil the thought behind it was. Because I’ve gotten it a number of times since then, usually phrased more euphemistically—I discussed it with the team and there were some concerns about the strategy of one-time rather than continuous dosing, that kind of thing—but the underlying sentiment has been the same.

“What if we made it worse, so that people had to buy more?”

Because it turns out the wingnuts out there ranting about They’re suppressing the cure for cancer, so that they can keep us hooked on toxic treatments!—they’re tapping into something, although it’s never that overtly sinister, right? It’s not executives meeting in a smoky back-room, putting out hits on researchers getting too close to the truth. The banal reality is that, in any domain, “cures” programs have a harder time getting funded. But the net effect on our society is the same, and it’s a manifestation of the same mentality, which is what I’d call pathogen mentality.

A pathogen takes as much as it can from its host—extracting maximum value at every turn. It does not live in stable equilibrium with the creatures around it, instead reproducing (you might say “scaling”) as rapidly as possible, producing toxins that get it coughed out, or shit out into the water supply so it can do the same thing to its next victim.

It’s a fantastically fit strategy. For a time.

But eventually, when your core strategy is to take without restraint, to take as much as you can, you’re liable to sicken and kill the larger organism that makes your existence possible. And that lands you the same place it lands the host: six feet under.

And this is why, of the thousands of species you find in the human microbiome, a vanishingly small fraction are overt pathogens. Because the most wildly virulent success, in that framework, ends in the host’s death—followed shortly by your own. A more restrained pathogen (take Giardia, for example) can do well enough for a while, by attenuating: just take enough to sicken the host, without killing them. But this creates a selection pressure: eventually a Giardia-proof human will show up, outcompete his sickly, susceptible cousins, and put the parasite out of a job.

And so my response to everyone who’s suggested engineering in a “suicide circuit”, or some kind of auxotrophy, such that you’d have to take a daily supplement in order to keep the cholesterol-reducer alive, has generally been a jovial: Fuck off, you cartoon villain.

Because that’s pathogen mentality. It’s trying to win at life, because you’ve failed to recognize that a “win” means game over, just as much as a loss. The goal of life, the only success to be had, is to keep the game going—and to that end there is only one strategy that works in the long run.

Symbiosis.

The only creatures that succeed in this long, long game of life are those that thrive by helping everything around them thrive. Taking what they need, yes, but giving back. Not more, but different. This is not a charity ethos; it’s rooted in the simple fact that one man’s trash is another man’s treasure.

Take fiber. What can we do with fiber on our own? Nothing! That’s part of why plants lock away all their carbohydrates in these indigestible polysaccharides—because, in the wild, something like sugarcane would be eaten to extinction in five minutes flat. But your gut bacteria look at cellulose and xylan, all the inert roughage in broccoli or oats, and go Oh, I’ll eat that! And the waste they produce, in the process: acetate, butyrate, the short-chain fatty acids—these are the grease that keeps the gears of your molecular machinery running smoothly. Bug shit! You can’t live without it!

Rumpelstiltskin

Everyone’s heard the story, right? The fairy tale about the woman who meets a little imp that can spin straw into gold?

Well I, too, am a little imp, and I have found a way to spin the cheapest part of the pig, a little bit of soup, and literal shit into that most priceless of commodities: human life. Quality and quantity of it. Time on this Earth.

We have a chance to give that to people. To sell it to them, at a very handsome premium! If that doesn’t already feel dirty enough—if you’re not interested unless you can extract rent on it? I don’t think I need your money.

And I think there’s still room in the world for this kind of thing, if we can resist the temptation of “shrink-flation”, of greed. After all, think about what happens when, the first time somebody orders a product from you, you deliver a minor miracle to their doorstep; something that adds ten years to their expected lifespan, because it drops their cholesterol by 65 points. That is how you create a lifelong evangelist, that is what we call organic reach. Now think about the value in creating a lifelong evangelist who is going to live an unusually long time. That is symbiosis.

And, y’know, inshallah we put ourselves out of a job because there’s no more diseases left to cure, but people are not going to stop taking antibiotics anytime in the foreseeable future: the average American takes them once a year, which we know means something like a 10% chance of losing your coprostanol-producing bacteria. And yes, that’s recurring revenue—but you know what’s going to happen, the second time a person has to come and pay us $5,000 to re-acquire this trait?

They’re gonna be a little more careful with it. And the next time they go to the doctor with an ear infection, and the doctor says “I’m writing you a scrip for erythromycin,” they’ll say “Wait a minute doc, don’t they make, like, antibiotic ear drops? Can we do the drops instead?”

And they do! And a lot of the time, you can. And so we won’t get a third $5,000 from that guy, but y’know what? Mazel tov.

Because, as I’ve written about before, taking any oral antibiotic increases your chances of ending up with MS, and depression, and Parkinson’s, and infertility—and now this guy’s risk of those conditions has not gone up, the way it would have if he had taken the drug. And so among our customers, just by creating this cognizance—there is more to you than the human—there will be less depression, and fewer suicides. The kind of fringe benefits that it’s hard to quantify, so which we call luck.

So now our lifelong evangelist is not just unusually long-lived, but unusually lucky. He’s happy, his skin is good, his kids are healthy. He ages gracefully, and doesn’t suffer the cognitive decline that sours the hearts and closes off the minds of some of his peers. And when he recommends this eye-wateringly expensive product to his neighbors, maybe the sour-faced ones will say to themselves Feh! Of course Mr. Lucky has $5,000 to blow on magic ice cream. But some of them will listen. Will recognize that, to really get lucky, you have to bet on yourself—to invest in yourself.

That is what we’re offering people: the ability to invest in themselves, in this screamingly literal way, and to actually own what they buy. Maybe not forever—but for as long as they can keep it alive. Maybe that’s a lifetime, maybe it’s several lifetimes—because bacteria don’t age the way we do. The strains we’re selling, they’re already heirlooms; they’ve been passed down for a thousand generations, and if you take care of them, eat right, avoid antibiotics—your kids will inherit them too, and their kids after them.

Life is the fruit of the orchard I am planting here, and money will grow on these trees regardless. And yes, little worms will come to nibble at those leaves: inefficiencies that could be optimized out. But let them be, and I think you’ll be delighted to find: they spin the most marvelous silk.

🖖🏼💩

“She’s in free fall”, my old friend says. Still all-there, mentally, but her motor neurons are flickering out, one after another. Months, if not weeks, ‘til the ones controlling her lung function go. It’s heartbreaking. It’s happening so fast.

She wants to know if there’s a gut microbiome angle on the disease, if there might be something we can do to slow it down—just to buy them some time.

I get these asks every so often. Every time, I take it dead seriously. Maybe it’s a vote of confidence, from the people who really believe in me and the work I’m doing—that they trust me enough to ask my advice on their loved one’s health. Maybe it’s just desperation; banging on the door of the witch doctor’s shack, once the more respectable physicians have started trotting out phrases like make her comfortable, complete with carefully practiced brow wrinkles, mouths set just-so.

Either way, I always try. Even if I don’t know much about a particular disease, I’ll do what I can to learn about it—do a lit dive to see if there have been studies of the gut microbiome in patients. Half the time, getting up to speed on the biology of a disease like ALS isn’t hard, because we still don’t know shit about fuck about what’s actually driving the disorder, in spite of decades of research.

Everyone wants to know if there’s anything we can do.

There usually is.

The Process: A Guided Tour

My first check is usually a literature search for any reports around fecal microbiota transplant (FMT) in a given disease, because it’s the most general test of microbiome involvement—whether by the presence of something nasty or the absence of something protective.1

A quick lit search turns up a small study—really just a pair of case reports—of FMT in ALS. Hey, I’ll take it. Both patients were hospitalized, trach’d at admission and dependent on mechanical ventilation to keep them alive. After FMT, both managed to wean off the ventilators and breathe independently again, at least for a while. Both also regained the ability to eat without choking. That’s pretty miraculous; typically once the doctors start stabbing holes in your neck, it’s all downhill.

So this is great! If we can get my friend’s mom an FMT, we might be able to help her. But, for some deeply stupid reasons, that’s easier said than done—at least if you want it done by a doctor. But maybe we can learn a little more about the disease from this FMT study, see if there’s anything we can do in the interim to buy us some time. Fortunately, they sequenced both of the patients’ microbiomes, as well as the donor’s, before and after the treatment.

Now, the most obvious thing that’s out of whack here is the abundance of the genus Enterococcus. In both cases, it’s the dominant component of their microbiome at baseline; about 40% in Case 1, and well over 50% in Case 2. This is an absolutely bonkers amount of Enterococcus to have in your gut. In the general population, you might find Enterococcus in the guts of 1 in 25 people, where it might make up something like 1% of their microbiome.

The fact that it’s so drastically elevated in these cases—and that it’s suppressed to near-zero after the FMT, concurrent with improvement in the course of the disease—has my hackles up.2

So: what do we know about Enterococcus and its biology? I wrack my brain. An ICU nurse that I used to know talked a lot about VRE—vancomycin-resistant Enterococcus, one of the newer and deadlier “nosocomial” pathogens. A problem you only get in the wake of a course of antibiotics, when all the native flora are gone, so that all it takes is a few stray cells with the right resistance genes to waltz in and set up shop. That’s probably how these ALS patients ended up like this—you only get 50% abundance of one or two species if everything else was completely nuked at some point in the not-too-distant past. Of course, antibiotic use is associated with ALS risk.

I shake my head. A few more facts rattle out. It’s in the same family as Lactobacillus and Streptococcus, meaning it’s oxygen-tolerant—part of what helps it spread in settings like hospitals. There’s a veterinary probiotic that uses some strain of Enterococcus, although whether that’s E. faecium or E. faecalis escapes me. I survey this paltry collection of puzzle pieces; certainly nothing that looks like it connects to a paralytic motor neurodegenerative disease. But I know very little, and Google Scholar knows a lot. So we search: “Enterococcus paralysis”. Sort of a long shot—but it turns up something interesting. It's a study in nematodes, describing an experiment to figure out what makes Enterococcus a sometimes-pathogen.3

See, most of the common pathogens are pretty businesslike about it: Vibrio cholerae produces a protein we call cholera toxin, which causes salts to flood out of your intestinal cells and into your GI tract, which in turn makes you shit your guts out.4 E. coli produces lipopolysaccharides that piss off your immune system, and the nastier varieties make something called shiga toxin, a protein that prevents your cells from making new proteins, which triggers the cellular seppuku known as apoptosis.

But Enterococcus has no such protein toxins, as far as we know. It doesn’t produce LPS like E. coli does, or eat the host’s cells from the inside, like Salmonella. So how does it cause disease?

Apparently: by producing hydrogen peroxide—H₂O₂.

The chemical is a household name, thanks to its role as an all-purpose disinfectant and hair bleach, but it also plays an important role in our biology; it’s produced inside your own cells by an enzyme called superoxide dismutase, or SOD.

SOD’s job in your cells is to clear out a highly reactive form of oxygen that would otherwise wreak havoc; it does this by grabbing it and converting it into hydrogen peroxide. But hydrogen peroxide isn’t exactly safe either (as you might have guessed from the fact that it’ll bleach your hair), so your cells have another enzyme called catalase, which scrubs out the H₂O₂ as it’s produced, converting it back into H₂O and regular O₂

So the body is equipped to handle hydrogen peroxide, but only up to a point—and if fully half of your gut microbiome is an H₂O₂-producing species, you’re almost certainly well past that point. (Recall that, in my longform piece on phenethylamines and tryptamines in schizophrenia, we did some napkin math using the pretty safe assumption that there’s at least 200g of bacteria in the average gut microbiome…implying that both of our ALS patients had something like 100g of Enterococcus in their guts, pre-FMT.)

So here we have a rudimentary disease model: another simple twist on the classic concept of auto-brewery syndrome. Last time, we swapped out the yeast for Ruminococcus gnavus or a similar species, and the alcohol for phenethylamine and tryptamine. This time, it’s an Enterococcus, and our molecular monster-of-the-week would be hydrogen peroxide.

Nerve Bleach

Does this track with what we know about ALS?

Well, I don’t know shit about ALS, but I do another quick search: “Hydrogen peroxide ALS”, and…huh, would you look at that! A whole review on the interaction between the two.

For background: ALS is like a number of other mystery disorders, in that—while there’s no gene mutation that’s necessary for the disease to emerge—there are some that seem to be sufficient. Autism is one of these disorders: mutations to a gene called SHANK3 produce a syndrome that more-or-less fits the diagnostic criteria for autism, but this explains only about 1% of cases—the vast majority of people with autism have SHANK3 genes that are perfectly fine. Even so, the outward similarities between the SHANK3-mutation kind of autism and the other kind suggest that maybe we can learn something about the biology of the “other 99%” flavor of autism by studying what the SHANK3 gene does.

In ALS, likewise, something like 90-95% of cases are sporadic—meaning there’s no family history and no clear genetic cause. But the other 5-10% of the time, the disease is inherited, with patterns that tend to follow a particular gene variant—a mutant version of superoxide dismutase, the enzyme that produces H₂O₂.

Now, oftentimes when we talk about disease-causing mutations, we’re talking about loss-of-function mutations, where a random change to the genetic code alters a protein in a way that breaks it. Think of albinism, for example: typically a loss-of-function mutation in tyrosinase, the gene that makes melanin. And this is sort of the natural, default way to think of mutations, because life is a very finely tuned thing; if you open up the binary of a computer program and start flipping bits, you stand a good chance of breaking it, while the odds that you’re going to make it work better are pretty much slim to none.

But it happens in biology all the time…more often than the coding analogy would suggest, because—while a computer program is designed by the coder to fulfill a particular function in an efficient way—protein structures are arrived at more or less by the process of “open up the binary and start flipping bits”, meaning there’s often a lot of room for improvement. And that appears to be what’s going on with the mutant version of superoxide dismutase that appears to drive some cases of ALS: a gain-of-function mutation that makes it a more efficient producer of hydrogen peroxide than the standard-issue version.

So—because I’m grossly under-informed and ignorant of most of the research that’s been done in this field—this all seems pretty pat to me. If familial ALS can be caused by a mutant version of SOD1 that spins out too much H₂O₂, having a microbiome that does the same thing seems like a solid explanation for the sporadic version of ALS.

Now, like last time, this hypothesis is more about a molecular pathway than an individual species—just as Ruminococcus gnavus, Enterocloster asparagiforme, and certain species of Blautia can all produce tryptamine and phenethylamine, there are a number of other bugs that produce H₂O₂ under the same conditions as Enterococcus. Lactobacillus and Streptococcus are among the best-known, so you wouldn’t necessarily expect Enterococcus to be the only microbiome feature associated with ALS.

This isn’t as well-fleshed-out a hypothesis as it could be; but the fact that—fully ignorant of the link between ALS and the mutant SOD1 gene—I found my way to “These people ought to have way too much hydrogen peroxide floating around”? That’s good enough for me, for now—because it might be good enough for my friend and her mom. “Enterococcus infection” is something any hospital care team knows how to handle; we’ve just got to convince them to send a fecal sample to the bacteriology lab, see if there’s an unusual amount of Enterococcus or something similar, and—if there is—find an antibiotic that it’s susceptible to. That’s not a long-term solution, especially since overuse of antibiotics is how we get things like VRE…but maybe it can buy us some time.

—🖖🏼💩

Now: in my last post I made some bold claims about the microbiome’s role in the etiology of schizophrenia. A few days later, prominent rationalist, psychiatrist, and Substack OG Scott Alexander happened to put up an ~8,000-word explainer on the problem of “missing heritability”, grappling with the fact that a lot of heritable traits like height, IQ, and—yes, mental illness—don’t map cleanly to the genome the way e.g. blue eyes does.

Between two or three people independently linking the piece in the comments of the “missing heritability” post, and me sharing it in his weekly “open thread” shortly thereafter, he finally decided to let us all know how he feels, putting out what one reader described as a “pretty damning takedown of [the] theory”.

I wouldn’t go quite that far, but he certainly didn’t mince words about disagreeing with me, and he gave his reasons why. And if there’s one thing I love about the rationalist worldview, it’s that it has no place for the notion of “agree to disagree”: If two perfectly rational people disagree, one of them is likely missing information the other has, or else is misinformed. And if, through argument, they can find the asymmetry in their knowledge-bases or worldviews, together they can get their facts straight, and whoever was wrong can change his stance.

In theory.

In practice, we're all human. Even a very rational person is susceptible to biases and fallacies like sunk cost. So when, over the course of your prolific career, you've dedicated an awful lot of words to exploring The Mystery of X—long explainers on why X is such a mystery, on its various aspects and manifestations, how we know it is not A, B, or any of the other letters down through W, and (just last week) “Eight Possible Explanations for the General Class of Phenomenon that X Seems to Belong To”…you get some inertia. And when some fresh-faced twat comes along and says “Oh, X? Not a mystery at all, that’s just well-understood-but-obscure thing Y”…it's hard to take him seriously, even if he's got a compelling argument.1

So the rebuttal is not unexpected. I might have preferred to explore these finer points in dialogue, rather than a format which feels a lot like skywriting…but I’m glad to at least see the guy’s objections outlined so I can address them, because they’re all eminently addressable.

As a warning: This is probably the longest substack post I’ve written, clocking in around 8k words. This is the inevitable fallout of making a post which proposes a grand unified theory of the biggest problem in the field, using fewer words than your average recipe blog SEO post. As noted at the top of the last piece, it was a transcript of a talk for which I was allotted a total of 10 minutes, delivered to an audience at a neuro-focused conference, so a certain amount of background was glossed over for the sake of brevity. I think the likes of Scott Alexander assumed that I had told as much of the story as there was to tell, and that he thus knew everything he needed to know in order to respond, when in fact I told the briefest possible version, because—unlike some people I could mention—I try not to make my readers slog through 8,000 words of gory details unless there’s a reason. But an idea this big necessarily takes a certain number of words to shore up and flesh out, so if you believed me about the five-minute version, you’re free to go if you like.

I. Discount of Monte Carlo

We’ll start off with Scott’s assertion that the following two statements are perfectly compatible:

-80% of variance in schizophrenia risk can be explained by the genome, and

-If your identical twin—who has the exact same genome as you—has schizophrenia, there is only a 30-40% chance that you will have schizophrenia.

(The term “concordance” will be used a lot below; this is what it refers to—when both members of a pair are the same in some way. Concordant=matchy, discordant=mismatch). Scott says:

That sounds wrong to me, but so did the Birthday Paradox thing until I saw the proofs on it, so I’m open to the notion that I don’t understand statistical variance, or that my intuition is the problem here. He goes on:

Okay but the thing is: No. You don’t. You fumble the one line of math involved, resulting in your simulation yielding a result within the (generous) target range you set for yourself. Then you get corrected on the math by a reader, implement the correction, find that the results no longer support your conclusion…and then you don’t change your conclusion, dismissing the discrepancy by gesturing at someone else’s work, which—according to commenters on your post—you have also misapprehended.

To break this down: Alexander is invoking this “simulation” as support for the plausibility of the claim that 80% of schizophrenia is genetic. The condition under which the simulation would support this claim is if it yields a twin concordance rate somewhere in the range of 0.15-0.5.

This is not a particularly tough mark to hit. Fully one-third of the entire range of possible outputs fall within it. You’d have pretty good luck just calling the “RAND()” function. But he still misses the broad side of the barn.

By way of explanation, he points out that 10% concordance is low out-of-range, so he’s not worried about it because any environmental similarities (same parents, born the same time of year) should increase it, and likely bring it up into the range of actual observed values. Obviously this is the point at which he should have said “hm, maybe clicking around in a spreadsheet full of randomly generated numbers is not a useful way to study a complex human disease.”

But now we find ourselves in a truly remarkable position, because according to Scott’s math: if 80% of schizophrenia variance is genetic, identical twins with no common environment should only have a concordance rate of 10%. So, say we go out in the real world and find ten schizophrenics who have identical twins. And we check the twins, and we find four of them also have schizophrenia—i.e. a concordance rate of 40%. Scott’s result implies that 3 of those 4 owe their matching disease status to things outside the genome.

So, for those keeping score at home: if the disease is 80% genetic, we’d expect 10% baseline concordance among identical twins, which is how we know that up to 75% of real-world twin concordance is due to things other than genes.

Does that sound right to you?

Maybe all this is technically possible under some weird definition of the words “risk”, or “genetic”, but if so, these definitions are so divorced from their commonsense meanings that it’s counterproductive to try and use them in conversation.2 I have to assume Scott isn’t just blatantly lying about the simulation result and hoping nobody clicks through, so either:

• he forgot about the correction and has that whole situation wrapped in a bundle labeled “evidence schizophrenia is 80% genetic” in his mind, or

• he used an LLM to write his rebuttal post, and this was sort of a “hallucination”.

The more I think about the latter possibility, the more likely it seems. It fits the pattern of confidently wrong assertion-with-citation that you get from current models, Scott just mentioned dealing with o3 hallucinations in the writing of a recent piece, and he seems like the kind of guy who would have a custom GPT trained on his corpus. I also thought “Contra Skolnick” came out suspiciously quickly after I posted in the Open Thread for the week, and figured he must have had it in the works since he got comments on the “Missing Heritability” piece…but that also fits with a timeline where he just had ChatGPT bang it out, looked it over once, and hit send. If this is the case, I will be absolutely furious, but also delighted that I got to live out a classic scifi trope.

So let’s move on, because all this is just the argument for why the genomic hypothesis shouldn’t be DQ’d from the get-go, on the basis of monozygotic twin concordance being below 50%. Centering the discussion on “Is the genomic explanation even mathematically possible, given what we know about genetics and statistics?” draws focus away from the real question at hand, which is: “How well does this hypotheses do at explaining what we see and what we know about this disease?”

So, just for fun: take the microbiome out of it. Imagine a miraculous sequencing breakthrough happens tomorrow, where they fix some long-overlooked flaw in the tech, and we suddenly find that—where most people’s genomes have only one copy of this aromatic amino acid decarboxylase gene—schizophrenics have, on average, 45 copies.

This would presumably be hailed by the Scott Alexanders of the world as rock-solid evidence that we’ve finally found the cause, or at least made a monumental breakthrough, because a 45x difference in a monoamine neurotransmitter-related enzyme blows every other gene correlated with schizophrenia right out of the fucking water, both in terms of signal strength and mechanistic plausibility. You’d still have to explain why it waits to kick in until a certain age, but we’ll get to that.

In terms of shedding light on the situation, the best the geneticists can do is five thousand mysterious SNPs, like stars in the night sky. There are a lot of them! And as they build better telescopes, they will discover more.

But they won’t make it any easier to see.

I offer you the sun.

There is just one. But it’s a good one.

II. A Loving Reminder to the Peanut Gallery

Scott’s “Contra Skolnick” opened by describing me as a “gut microbiome expert”, a label which I’ll wear with pride—but it sparked a few sniggering comments to the effect of “Mandy Rice-Davies Applies”, i.e. “Of course, as a gut microbiome scientist, he’s going to insist that schizophrenia is in the gut microbiome.” And while I mentioned this in the reply to those comments, I felt it worth reiterating here: I was not always a poop scientist. My undergrad was in physics, and I didn’t go to grad school; I had to hustle and muscle my way into this field, making risky moves and some real sacrifices, leaving a nice job as a science writer at APS for the chance to become…a turd-wrangler.

I chose to do so because, when the key facts about the microbiome—that it is at once both highly heritable and wildly variable—started to percolate into the public consciousness a decade ago, it became apparent to me that this could be as important for our understanding of biology as Mendelian inheritance was, and that it’s a promising potential explanation for many of the chronic illnesses that thus far have stumped our best and brightest.

This doesn’t mean I’m free of biases—if anything, I’ve got a lot more sunk-cost than someone who happened to be a microbiology major graduating in 2014. Worse, I have always been a relentless optimist, straining to see any way in which diseases like schizophrenia might not be hard-coded, but rather mutable, and therefore fixable. But I’m an empiricist, and I pride myself on that. I’m aware of my biases, and I go to great lengths to avoid letting my desire for something to be true impact my assessment of whether or not it is. Obesity is a great example of something where it would be nice, but I don't think we’re going to be able to treat it effectively with microbiome modifications like FMT, because a number of people have tried and failed.3 Point is: Experiment is king, and if getting rid of the tryptamine-producers in a schizophrenic person’s gut doesn't bring and keep them out of psychosis, then hey, I’ll happily take this hypothesis out behind the woodshed.

I would be quite surprised if it didn’t help, because—as I pointed out last post—a meta-analysis of metabolomics studies has found that the correlation between the abundance of Ruminococcus gnavus in the gut and the concentration of tryptamine and phenethylamine in the bloodstream is the strongest correlation between any microbiome feature and any serum metabolite, period. That is to say: if we know literally one thing about the gut-brain axis, it’s that having a lot of gnavus and Enterocloster in your gut microbiome means having a lot of tryptamine and phenethylamine in your bloodstream (and thus your brain)4 Maybe removing them wouldn’t do anything, because it turns out these compounds have nothing to do with the pathology of schizophrenia! But it would be a very weird coincidence.

So, apologies for the long digression here, but I wanted to directly address the notion that I have conveniently formulated some theories wherein my expertise places me at the center of the universe. I saw the center from afar, and I have moved to it: Anyone who knows me personally will tell you that I have been yammering on about how many of these problems look like nails, even back ten years ago when the nearest hammer was located across a river of shit, and up a high cliff (also of shit).

The other reason I bring up my time as a physicist is because one of the arguments in Scott's rebuttal—the first paragraph, below—reminds me of the racket that the particle theoreticians have going on.

The appeal to our technological limitations is a powerful one: the supersymmetry guys and string theorists love to propose a grand unified theory, an ultimate explanation for all that is…which conveniently can’t be tested without a particle accelerator the size of Moldova, and which would cost the entire GDP of Moldova to build. So, tenure secured on the strength of a prestigious and effectively unfalsifiable theory, the profs can get back to theorizing about the existence of glueballs5 and teaching the occasional cosmology class, to ensure that the next generation of drone engineers is a well-rounded bunch. On the off chance that some government or coalition comes along that’s crazy enough to actually BUILD the damn thing, they can look forward to consulting on the construction, but rest easy knowing they'll be well into retirement by the time the beamline is operational and their ideas risk being put to the test.

All this to say: I would love to see the day where a new breakthrough in sequencing tech turns out to be biology’s Great Moldovan Hypercollider, but I'm not holding my breath, because this is the same spiel we've been hearing since the earliest days of gene sequencing.

On the second paragraph in the pull quote above:

I agree, it would be surprising if everything from height to educational attainment were determined by gut microbes. I don’t consider that very likely. But the unlikelihood of every such trait being gut-bacterial has no bearing on the likelihood that any one such trait should be.

Laid out in formal logic, the argument Scott seems to be making is:
Given: schizophrenia is a typical part of the set of “traits with substantial missing heritability”.
If: Schizophrenia is gut-bacterial in origin
Then: The rest of that set, e.g. IQ and height, would also be gut-bacterial in origin.
Given: it’s highly unlikely for IQ, height, etc. to be gut-bacterial in origin.
Therefore: Schizophrenia is unlikely to be gut-bacterial in origin.

Laid out like that, hopefully it’s clear that the if-then proposition involved is a classic squares-and-rectangles misunderstanding. This is a weirdly basic fallacy that I wouldn’t expect from a central figure in a movement that prides itself on logical acumen. Maybe more support for P(o3), or just a sign that Scott’s realized he can deploy long explainer articles like smoke bombs, and that it doesn’t really matter what he says after that, as long as it sounds punchy.

III. What Schizophrenia Does and Does Not Resemble

One of the main problems facing the field is that, if schizophrenia is mostly genetic, we need to be able to explain why mutations that are present in the genome from birth only start causing trouble once a person is basically fully grown. This is why phrases like “gene-environment interactions” and “two-hit model” get a lot of play in discussions of how the pathology of the disease might work. As Scott puts it:

And I’m glad he asked that question, because it’s one of the greatest strengths of a microbiome-centric explanation, not only for schizophrenia, but for practically any chronic and noncommunicable disease where a person is fine for most of their life, and then it seems like a switch just “flips” one day, and now it can’t be un-flipped.

Because, while your microbiome ordinarily exists in a pretty stable state of equilibrium, it’s still the most labile component of your biology. The relative abundances of various bugs wax and wane from day to day, from poop to poop—depending particularly on what kind of meal those poops started out as. But provided you don’t swing from full veganism to a carnivore diet overnight, all the species that are there one day will likely be there a week later, at roughly the same proportions.

But say you get norovirus, and spend three days straight shitting your guts out—or you get Lyme disease, and have to spend a month on some very heavy-duty antibiotics. This is the kind of ecological upheaval that can tip the ecosystem into a new equilibrium: just as stable as the previous one, perhaps even more stable, but not for the better. Maybe your Phascolarctobacterium has gone extinct, and then suddenly you’ve got three times as much Dialister, because those are basically the only two genera that eat can eat succinic acid, and your Bacteroides produce a lot of that. And then, over the course of the following month, you suddenly come down with a mysterious disease that has a stupid fucking name like “ankylosing spondylitis”, and which—for some strange reason—comes with a free side of IBS.

This is where the Chronic Lymers and the Branch-COVIDians come from: You get sick for a week and get over it, but you’re never the same after. In the case of norovirus (and maybe even COVID) this is sort of a consequence of the disease itself, but with Lyme it’s more the collateral damage from the treatment: burning down the jungle that is your gut ecosystem. Things will grow back, but not quite the same as before. And if suddenly you find yourself with too much bamboo, good luck, because burning it all down again is probably just going to get you even more bamboo.

So, in light of this, think about what we know about Ruminococcus gnavus’s prevalence and abundance in the human gut. It’s detectable in about 50% of people’s fecal samples globally. (Notably: those people have 33x the odds ratio of cardiovascular disease vs. gnavus-negative, if I’m reading this study right [I can’t be reading that right; someone check me on that?])

So it’s prevalent: a lot of people have it. But how abundant is it? How much do they have?

For the most part, not a lot; it’s typically less than 1% of the bacteria in your gut—and usually more like 0.1% or 0.3%

So this is how I picture it going: Say you’re in the gnavus+ half the population (there’s two kinds of people in this world, huh?), but you’ve got a moderately low relative abundance; maybe you’re the average American in that chart above, and gnavus makes up 0.01% of your microbiome. That’s a very minor component! For comparison, one species of Bacteroides—which I often describe as the trees of the forest—might be 10% of a person’s microbiome by cell count. But just to get a rough physical sense of how much this is: say there’s 200g of bacteria in your colon. 0.01% of that is 20 miligrams of cells.

Now say you take an antibiotic, and it happens to be one that your strain of gnavus is resistant to. And after that antibiotic, you find yourself in the white bar for the US on that graph: top quantile, with its low end at 1% relative abundance, and its top at about 10%. Now you’ve got between 2g and 20g of gnavus cells in your gut.

20 grams! Y’know what else weighs 20 grams? A prostate! A thyroid! And so suddenly, at some point in young adulthood, with a little bad luck and a class of drug that the average American takes once a year, you’ve gone from being the kid who gets a little weird when he eats too much pizza to being the guy who has grown a small organ dedicated to churning out psychoactive chemicals. A month or so after you’re off the antibiotic, your microbiome has reached its new equilibrium, and the switch has been flipped.6

In my parents’ day, the popular explanation for this sudden switch-flip was “hormones”; maybe something about the end of puberty activates it. And maybe “adolescent synaptic pruning” is the more evolved version of that, incorporating the fact that most of the schizophrenia risk genes worth talking about are immune system genes, and synaptic pruning has common heritage with some immune processes.
“If your microglia did a little too much pruning in the executive function department”, the thinking goes, “and then pruned the brake lines on the dopamine system…” you could get something that might reasonably be expected to look like schizophrenia.

But that still leaves a lot of explaining to do: how does that gene actually cause the bad pruning to happen, and why doesn’t it happen in the other 98.8% of the people who carry the gene? Or if it’s polygenic, how does that gene interact with any of the other genes involved to produce that effect? Because my main problem with the genetic hypothesis of schizophrenia is that practically all the risk genes identified to date…

(A) do so little to impact disease risk that you need a cohort of about 2,000 people to even identify them as being more common in schizophrenia, or

(B) have nothing to do with neurobiology, in terms of their known functions.

…if not both.

Saying “highly polygenic” or “gene-environment interactions” is not an explanation, it’s a description of an explanation that you hope to come up with someday. That kind of thing has slid by, thus far, because everybody at the table in these conversations is just as naked and perplexed as everyone else, a state that generally engenders polite agreement when the subject of fashion comes up. But the minute someone shows up with a simple, mechanistically valid, testable hypothesis and biomarkers to go with it? You gotta engage with the specifics of the idea!

If you want to argue that the effects of chronic phenethylamine and tryptamine infusion don’t resemble schizophrenia very much, be my guest—but then you have to contend with the fact that amphetamine psychosis is one of the only animal models of schizophrenia with any translational validity, meaning: the drugs that work to alleviate amphetamine psychosis also tend to work for the positive symptoms of schizophrenia.

In the comments, someone also raised that the hallucinations of schizophrenia are not much at all like those induced by serotonergic hallucinogens. Now, I have taken my fair share of tryptamines and phenethylamines in my day (arguably more than my fair share), and I think I agree with this person. At anything less than “breakthrough” doses, you don’t see things or hear things as much as you feel things, like a psychic high-pressure front that infuses every situation with a sense of ambient significance. As I understand it, the delusions and hallucinations of schizophrenia are much more like what happens to you after a few days of sleep deprivation: you glimpse the coat rack out of the corner of your eye and it makes you jump, because for a second you could’ve sworn it was someone standing there—and even after you calm down, you can’t shake the feeling of another presence there in the room with you. Or you’re dozing, and then startle awake at a noise out on the street—was that someone saying your name?

So consider that, if you put someone on a phenethylamine slow-drip, you’d expect them to sleep pretty poorly, if at all. Add in that feeling of impending…something you get from a tryptamine, and that sounds like a recipe for a pretty psycho time.

But I have never had schizophrenia—and neither, I suspect, has Scott Alexander. And because I don’t plan on Barry Marshalling this one unless I absolutely have to, I’d be curious to hear if any of our readers check both of those boxes, and can give us an actual firsthand opinion on to what extent one of those things is like the other.

IIIa. Flipping and Unflipping the Switch

Before we move on, this seems like as good a place as any to address one of the more thoughtful critiques that came up in the comments of “Contra Skolnick”, courtesy of a user by the name of Xcm, who asked pretty insistently: “Shouldn't a course of antibiotics to which R. gnavus is susceptible immediately result in noticeable improvement?”

And it’s a good question! But “we’d have noticed by now” is the kind of argument that often collapses immediately when you try to move it from the vacuum of imagination into the real world, with all its confounders.

Are we talking about a macrolide, or a beta-lactam antibiotic? Is it given by mouth, or IV infusion?

By mouth, the first tablet you give will maybe take 30 hours to reach the rectum, which is where it matters. If it's a protein synthesis inhibitor or something like flagyl, it won't do much to inhibit the activity of an enzyme that's already been produced, so you're probably waiting at least 2 defecations (somewhere between 25 and 80 hours) before things get any better.

Would we notice?

Well, was the person in active psychosis? That's when you'd expect an immediate improvement to be most noticeable.

But how often does a doctor prescribe an antibiotic alone to a schizophrenic patient in active psychosis, without also prescribing something to try and manage the psychosis? And if that thing seems to work, don't you keep them on it?

If they weren't in psychosis when you gave them the antibiotic, then yeah, maybe they see a durable remission in negative symptoms over the following months. So then we’re back to: why were we giving them the antibiotic? Say it's a UTI, and at 6 months post-drug, they say "hey doc, I've been feeling a lot better, no like a LOT better since then..." How much better do they need to be, and for how long, before you to write up a case report rather than saying "I'm very glad!" and chalking it up to clearing up the UTI?

The problem is that, in the short term, remission is nothing newsworthy, and in the long-term, bacteria grow back unless you take such a complete scorched-earth approach that it's liable to cause problems of its own. If you nuke it all and then replace it with an FMT, that’s one thing—but as far as I can tell, nobody has tried an antibiotic+FMT for schizophrenia. The one scientific case report I can find, of a person making a complete and durable recovery from serious schizophrenia with no antipsychotics, is the story of a young man who got a bone marrow transplant. The protocol for this involves so much chemotherapy that it literally hollows out your bones, followed by a whole new immune system and thirty days in a sterile isolation room, during which time the man’s psychosis disappeared, and stayed gone for at least four years. And because the chemo drugs involved in that kind of thing are toxic to anything that has DNA (i.e. everything), this is one of the few people who I’m confident actually got a fresh start on a microbiome. Confounders abound, of course, but we can at least be pretty sure that the bone marrow transplant didn’t un-prune his adolescent synapses, so there’s that.

Also worth noting: that paper was really hard to find, because—if you put “antibiotics” and “psychosis” into the same query on scholar.google.com, most of your first five pages of results is full of case reports, studies, and reviews about the problem of “antibiomania”—the fact that full-blown psychosis is a relatively rare side effect of nearly every kind of antibiotic.

Remarkably, this parallels the pattern in multiple sclerosis, where practically all antibiotics increase your risk, but the risk ratio depends really heavily on the type of antibiotic. It would be one thing if it were specific to a class of drug: aminoglycosides are famous for causing hearing problems, and in that case it’s something about their particular mechanism of action—their molecular structure, and the way it kills bacteria—that makes them liable to do collateral damage to the sensitive cells of the inner ear.

But most of the different antibiotics in that table look nothing alike structurally, and they work by totally different mechanisms of action. The only thing most of them have in common is that they kill bacteria.

And I didn’t even think to bring it up in my discussions with Xcm, but it’s really a solid corollary to their question. WOULD we have noticed by now if it were possible to treat psychosis with antibiotics? I don’t know. But we’ve definitely noticed that it’s possible to induce it.

Subscribe now

IIIB. On Timing, and Aioli Problems

You ever had a really amazing mayonnaise?

Of course not. What would that even entail? Maybe there’s garlic mixed in, some spices? Well, that’s no longer mayo, is it? What you have there is an aioli, maybe a remoulade.

And I know it’s silly, but I think it gets to the heart of something that everyone knows is a problem in psychiatry, i.e. that the names we assign things can lead us to treat two things as the same when they’re quite different, or vice versa. I once saw a talk where an fMRI guy claimed that, if you break it down by what’s going on in the brain on their scans, the disease we call “major depressive disorder” could be as many as 12 separate phenomena.

But it goes the other way, too. One of the great mysteries of schizophrenia is its timing: why does it only seem to strike, as Scott put it, at ages 18-30? How do I explain that with bacteria?

Of course, it doesn’t. Just in my own life, that’s plain: two good friends of mine have siblings with it. One “grew into” her diagnosis: it started out as severe ADD and a learning disability at a young age; once she was old enough, she was transitioned to a diagnosis of schizophrenia without much fanfare or change in her medication regimen. The other got his diagnosis pretty young, compared to most: 13 or 14. Weirdly enough, that same friend’s mom had a bizarre amnestic episode in recent years, and apparently lost all memory of roughly the last decade. A stroke? Early onset dementia? We’re not sure. She was always the kooky mom, in our friend-group. Great fun, the perfect picture of “only slightly weird”, to use Alexander’s phrase.

Point being: Imagine for a moment it’s all as simple as I say it is, and schizophrenia is basically like being put on a semi-constant internal slow drip of psychoactive drugs, with the bonus that tryptamine appears to mildly fuck with your ability to synthesize proteins, because it looks too much like tryptophan. Now imagine that same process starting at age 2 or 3, while you’re still developing language skills. I mean, you can’t really imagine such a thing, because the brain of a child that age is so different from an adult’s, but we might expect them to stop learning new words. It might look a bit like some cases of ASD.

Or imagine it happening at age 70, when you’ve already got a full lifetime of experience navigating the world, the strong network of social bonds that younger-onset schizophrenics often don’t develop. What would that look like?

IV. Spouses, Siblings, Species, and Strains

One of Scott’s main objections to this hypothesis was that schizophrenia’s heritability patterns don’t look like what you’d expect from a microbiome-driven disorder—or rather, they don’t look like what he’d expect, because he’s operating under some false assumptions.

Here again, this is my fault for omitting some details in the name of brevity. Because, while the microbiome is certainly kept on a longer leash than most of the rest of our biology, the leash is still there—and in a lot of ways it does lead back to the genome. The main thing holding the other end of that leash is the immune system, but there are also major contributions from less-direct factors like mucins: amino-sugar-rich proteins secreted by your intestinal cells, which serve as both a barrier and a food source for certain bugs. Or think about lactose intolerance, or even the olfactory receptor genes that determine your sense of smell, your taste preferences. To the extent that these differ between fraternal twins, you’d expect to see more microbiome differences between them than you do with identical twins.

And if you plug some basic keywords (“microbiome twins identical fraternal”) into google Scholar, you’ll find that identical twins do, in fact, have more similar microbiomes than fraternal ones.

What’s especially neat about the above table is that they’ve even broken out the analysis phylogenetically, so we can see that—while monozygotic (identical) twins aren’t more similar than dizygotic (fraternal) twins with respect to the Bacteroidaceae in their microbiomes, they are more similar with respect to the Lachnospiraceae—which happens to be where Ruminococcus gnavus sits. Of course, that’s only by the “weighted UniFrac” metric, and I can hear the skeptics already: What, he’s just gonna pick the metric where there’s a significant difference in the taxon of interest? What about the unweighted UniFrac?

Well, first things first: UniFrac distance is basically a measure of “How different are these two microbiomes?” (UniFrac stands for Unique Fraction of the tree of life) but that’s a tricky thing to reduce to a single number, so you have to tread carefully when trying to figure out what a study like this is actually saying. Is a pine forest in Oregon more like an oak grove in Oregon, or is it more like a pine forest in Japan? It depends on what you really want to know about, but it’s worth learning how UniFrac works so we can at least talk intelligently about these findings.

For a simple case, imagine two valleys next to each other, and let’s say that they both have the same two species of tree in them: one oak and one pine. That’s a UniFrac distance of 0, because there’s nothing unique between them. But if we introduce squirrels to one valley and not the other, now there’s some UniFrac distance between them. Let’s arbitrarily say that’s a distance of 1. If we put a couple foxes in the other valley, the UniFrac distance between the two would go down, maybe to 0.5, because—while foxes aren’t that much like squirrels—they’re more squirrely than trees, which were the closest thing in the valley before the fox showed up. If, instead of foxes, we put an eagle in the other valley, the UniFrac distance would still go down, but not by as much: a bird is more like a squirrel than a tree, but less like squirrels than a fox. A chipmunk, on the other hand, is about as squirrely as you can get without actually being a squirrel. So if the two valleys both have the same species of trees, but one has squirrels and the other has chipmunks, we’d maybe have a UniFrac distance of 0.1.

So now take it back to the two-trees-only case, pines and oaks in each of these valleys, UniFrac distance 0. But say one valley is a 50/50 mix, while the other valley is all pines, except for one lonely oak. Now it doesn’t seem right to say “there’s no difference between these two,” right? That’s why weighted UniFrac also takes the relative abundance of the species into account, rather than just checking “what’s there?”

Knowing all this and taking it back to the table above, we can now see it’s telling us that—where the Lachnospiraceae are concerned—identical twins have more similar microbiomes than fraternal twins, when you give due weight to the ones that actually make up a significant portion of the ecosystem.

A beta-diversity metric like UniFrac isn’t really suited to investigating a hypothesis centered on a single gene or enzyme, like this is—it’s the kind of thing you look at when you don’t know what you’re looking for, which is why the points Scott makes below about spouses don’t really hold up. But it’s still a neat thing that what Scott thought was a “gotcha” turns out to supports the hypothesis.

Okay, to quickly address that last one:

Can we think of any reason why there might not be a lot of studies showing that children adopted out from mentally healthy parents to parents with schizophrenia acquire the schizophrenia risk of their adoptive families? Any reason at all.

This is the kind of objection that tells me that the guy gave just enough thought to my words to write a semi-coherent rebuttal, and even less thought to his own words as he wrote it. And this is the thing that pissed me off enough to elicit such a long and thorough reply.

It is right to be skeptical. It is a sin to give your skepticism the benefit of the doubt.

So let’s talk about the “epicycle” he describes in the first bit, where he’d expect the genes that determine your microbiome to show up on GWAS as schizophrenia genes.

If you’ve been reading attentively this section, you’ve maybe already put it together: I mentioned in the bit on synaptic pruning—practically all the major schizophrenia risk genes are immune genes! Mostly located in a region called the Major Histocompatibility Complex, or MHC. This is part of how the body differentiates its own tissues from those of other species which might be invaders. It’s not fully understood how the body compiles or enforces its list of which bacteria are friends vs. foe, but it’s all too easy to imagine that one of the genes in there is the molecular equivalent of a “DO NOT ALLOW ON PREMISES” poster, with a little picture of his glucorhamnan surface polysaccharide.

V. The Poison Makes the Dose

Probably the most consequential objection to this theory is the possibility that we've got the direction of causality wrong here. All the schizophrenic people in this study were on medications like clozapine, so how do we know that clozapine doesn't cause the observed microbiome differences?

And it’s true, the authors of the paper note that there’s a significant correlation between antipsychotic dosage and the abundance of bugs like R. gnavus, and even go so far as to conclude that it's probably the medication driving the observed microbiome differences—so I get why it seems shady that I didn’t mention the authors’ initial conclusion in my post.

The possibility that clozapine impacts the microbiome is worth thinking about, especially in light of the fact that it's known to produce constipation and other pretty serious GI symptoms. And, if the participants in this study had been randomly assigned to various clozapine dosages, then sure, the association between drug dose and gnavus abundance would be pretty solid proof that the drug drives the bug.

But that’s obviously not how this works: a patient’s medication dosage is a product of symptom severity—the more severe their schizophrenia, the more antipsychotic it takes to keep them out of psychosis. So there are two models here: in one, the symptom severity drives medication dosage, and medication dosage drives gnavus abundance. In the other, gnavus abundance drives symptom severity, and symptom severity drives medication dosage.

If you’re not a total shit-head like me, I can see why Model 1 would seem like the more likely possibility. Maybe you look at the top few rows in your results table, and go “huh, what is Faecalicatena gnavus, and is there any way it could drive symptoms?” and you do a lit search on it. Unfortunately, bacterial taxonomy is a fucking mess, so you’re probably not going to find the paper from 2014 where they characterized its aromatic amino acid decarboxylase, or this more recent deep-dive on the species’ pangenome, where they show that it’s a core feature of the species—because in both of those it’s referred to mainly as Ruminococcus gnavus. You’re also not going to be aware that Enterocloster—one of the other taxa most enriched in the schizophrenic gut—is one of the only other gut bacteria which possess such an enzyme; I only know this because we were looking into this pathway back at Holobiome and I read Fischbach’s original 2014 paper.7

But this is a key fact, because in light of it, it suddenly doesn’t seem like nearly as much of a stretch to suggest that the abundance of the bacterium is what drives the medication dosage. And if you want to postulate that it goes the other way around, you find yourself going “guess it’s a weird coincidence that a schizophrenia drug happens to facilitate the growth of a bunch of tryptamine- and phenethylamine-producing gut bacteria”. (Not THAT weird, because monoamines encourage motility, which probably contributes to a low baseline relative abundance of gnavus in people who aren’t otherwise constipated—but still.) Like synaptic pruning, it’s the kind of thing that makes sense to postulate in the absence of an explanation involving fewer unknown steps, but in light of what we know now, falls away to Occam’s razor pretty quickly.

VI. The Ball

If I may speak directly to you, Scott: I get why your first instinct was mostly to defend your framework rather than engaging earnestly with mine—to dismiss me on something like the MZ/DZ twins concordance before even checking whether it was true.

Because this game has been going on for centuries without a single point scored. When was the last time anyone even saw the fucking ball? For our entire lifetimes, it feels like everyone’s just been milling around the field looking for it, speculating about where we might find it, dreaming up plays for when we do. Does anyone remember what it looks like? How sure are we that there even is one?

So I understand the skepticism, when you hear someone shout—not too far off—I’VE GOT THE BALL!.

The false starts / true starts ratio, to-date, is ∞. And when you look over, and you see that the ball in question is one of these…

…you roll your eyes. This is a joke, right?

Yes, absolutely. If there’s one thing my artist friends are pretty clear on, it’s that all of this—the whole ballgame, you and me included—none of it makes sense except as a cosmic joke.

But that just would be the punchline, wouldn’t it? The one thing nobody’s tried, because until this very Sunday morning, it sounded…well, yes, crazy.

So I’m telling you: I’ve got the ball. I’m running it.

And if I can make it to the end zone, I guess we’ll see what happens. But we’re on the same team here, unless you’re playing for Moloch, so the only question is: Do you want to sit on the sidelines and watch, or do you want to get downfield so I can pass to you?

Because I am pretty good, but I’m only one guy. I need help. Someone with an MD. Preferably a psychiatrist. Because, unlike the thousand SNPs in the genome, this is an extremely testable hypothesis: a course of vancomycin and rebyota. Drugs that you could write a scrip for tomorrow.

So this is all I ask you: adopt this hypothesis for a minute; hold it in your mind as if it were your own. I do not mean be gentle with it; you must be violent! The whole point is that you try to break it: smash it, chip at it, find the seams and pry them open. Yes, ask the questions—If that were so, then wouldn’t we see…—but then you must actually try and find the answers! Dig into the metabolomics table, read the papers, check your assumptions—see it for yourself. And if you don’t just toss it away, saying “I’ve seen a dozen like this”, I think you’ll be surprised. Nothing I’ve tried yet has cracked it, which is why I’ve invited the world in to come and have a kick. And if it crumbles in your hands, I’ll thank you. We didn’t want that one anyway.

Because we are after the truth.

And truth is indestructible.

—🖖🏼💩

Huge thanks to everyone who engaged in the discussion around this in the comments of my last post and “Contra Skolnick”. A lot of new readers came from that one, a lot of new ideas. It’s a heartening reminder that—even if it turns out Scott’s now a replicant—the best thing about SSC/ACX has always been the community of people who gravitated toward his writing, because they recognized the spark in it. Y’all know what the real thing looks like.

Now, if you know anything about the genetics of schizophrenia, I’m sure you’re thinking to yourself: “Bull shit.”

Because of course, if you know anything about the genetics of schizophrenia, you know: there is no schizophrenia gene! This is the great puzzle in the field, right? Because the disease is very clearly heritable: Having close relatives with schizophrenia is the strongest predictor that you’ll have schizophrenia yourself. If one of your parents has it, there’s about a 13% chance you will, too—more than 40 times the risk of someone whose parents are both mentally healthy.

But it’s squirrelly, somehow. Most people who get a diagnosis of schizophrenia are the first ones in their families; it sort of crops up out of nowhere most of the time. If you have an identical twin and they have schizophrenia, the odds are pretty high—30 or 40%—that you will too. But that’s for two people who have the exact same genome—which means that most of what determines whether or not you get this disease is non-genetic.

And researchers have run GWAS after GWAS in an effort to find the genes responsible, and time and again they come up with…scraps. Dregs! 1% here, half a percent there. This paper from 2013—one of the more recent big efforts1 estimated that, if you look at 8,300 different spots in the genome, you can account for up to 32% of risk.

This is, of course, the definition of an overfitted model. With 8300 parameters, you could probably build a genome-based classifier that tells you whether or not someone is named “Karen” with 32% accuracy.2 And while nobody could argue with a straight face that Karen-ness is genetic, the field of biological psychiatry is still largely shackled to the idea that schizophrenia is just “highly polygenic”, involving so many different genes, each with such a small effect size, that there’s no hope of figuring out how it works until we have a billion schizophrenics whose genomes we can study.3

So how do we square this? How do we explain something that’s clearly heritable, clearly biological, and clearly not-just-genetic?

Well, ten minutes isn’t a lot of time, so I’ll cut the theatrics. It’s in the gut microbiome.

See, many of the bacteria in your gut right now are strains that you picked up from your parents, back when you were still crawling around on the ground and putting everything in your mouth, picking your nose and eating it, and so on.

Those bacteria have stayed with you most of your life, eating the food you eat, reproducing faster than you can shit them out. If you have kids, you will pass those same heirloom strains on to them.

Heritable. Biological. But not genetic.

So I am not bullshitting you when I say I have found the gene responsible for schizophrenia. It’s just not a human gene.

One Gene.

See, in 2023, some researchers finally took a good look at the gut microbiome in schizophrenia, and they found that a bacterium called Ruminococcus gnavus is—on average—about forty-five times more abundant in the guts of schizophrenics than in mentally healthy controls. If you look just at treatment-resistant schizophrenia, it is over one-hundred times more abundant.

This would be remarkable in and of itself. But what’s even more remarkable—and what the authors of that paper did not seem to know—is that one of the defining features of this species, Ruminococcus gnavus, is that it possesses a promiscuous aromatic amino acid decarboxylase enzyme.

Now, my PI back at MIT used to object to the use of the word “promiscuous” in this context. She would say: I prefer “versatile”, when referring to an enzyme that can accept multiple substrates, and while ordinarily, fine, cute, I agree—in this case we are absolutely slut-shaming this enzyme. It’s got a real loose substrate-binding pocket and we are all the worse for it.

See, it takes tryptophan in the gut and turns it into tryptamine, and takes phenylalanine and turns it into phenethylamine.

Now, I love this crowd because, if this were a microbiology conference, I would have to spend the remainder of my time explaining what that means—but I expect everyone here has experience with tryptamines and phenethylamines, either on a personal or professional level.

But that means you probably know: if you just eat tryptamine or phenethylamine, they won’t do much to you. They get destroyed by monoamine oxidase (MAO) as they pass through the liver on their way into the bloodstream. This is why ayahuasca has two key botanical ingredients: one contains the hallucinogenic tryptamines like DMT, and the other contains a beta-carboline—a natural MAO inhibitor to run interference with your liver. Likewise, if you wanted an orally active phenethylamine, you might modify it chemically, attach a methyl group on the alpha carbon to protect it from MAO—and then you’d have alpha-methyl-phenethylamine, or amphetamine for short.

But because this is not a microbiology crowd, I want to clarify something: When I say “gut bacteria”, I am not talking about your stomach, I am not talking about your thirty feet of small intestine. The overwhelming majority of the microbial metabolic activity in your GI tract is happening in the final stretch, the large intestine and colon.

And what is so special about the colon? Chemicals absorbed in the colon skip first-pass metabolism! This is why the most truly dedicated degenerates boof their drugs, right? It’s a direct line into the bloodstream, for metabolites produced by your gut bacteria—more akin to smoking or injecting a compound than eating it.

Subscribe now

Now, I want to be clear: this is not just theoretical. I have personally reanalyzed the data from a published meta-analysis of ten different metabolomics/metagenomics papers—that is, looking at associations between the chemicals in the bloodstream and the bacteria in the gut, and throwing out any signals that didn’t replicate across at least three independent studies.

And in this meta-analysis, the strongest association between any microbiome feature and any serum metabolites—the very top of the thirty-thousand row table in their supplementals—is the correlation between the amount of Ruminococcus gnavus in your gut and the amount of tryptamine and phenethylamine in your bloodstream.4 The more of this bug, the more of these chemicals.

Now, tryptamine is a full agonist at the 5HT2A receptor—that’s the trippy one. Nanomolar affinity. Phenethylamine acts a lot like amphetamine, which is known to induce delusions, hallucinations, paranoia, and psychosis after prolonged exposure. In fact, this is one of the better ways to model psychosis in animals, if you need to test out a new antipsychotic drug; dose ‘em up with enough speed that they don’t sleep for a few days, and pretty soon they display a hallucinatory phenotype that can be reversed by dopamine antagonists.

But just as interesting as the effects of the compounds themselves is the body’s response to them. Phenethylamine induces neuroinflammation, so in response to it your brain up-regulates expression of monoamine oxidase B, to clear out what the liver couldn’t handle. And this does a pretty good job of getting rid of the phenethylamine, but in the process it also gets rid of serotonin, dopamine, and norepinephrine.

And since we all know what happens when you’re short on your monoamine neurotransmitters, you’d expect this to lead to a syndrome of anhedonia, avolition, flat affect, depression—in other words, the negative symptoms of schizophrenia…punctuated by episodes of delusions, hallucinations, and psychosis—the disease’s so-called positive symptoms—whenever the rate of tryptamine and phenethylamine production overwhelms the body’s capacity to remove it.

So there you have it, ladies and gentlemen: Schizophrenia in seven minutes or less. One gene! And even if you’re not totally convinced, you have to admit, it's a lot cleaner than “look at these 8300 SNPs, most of which are noncoding”.

Even so, there are a few more pieces to the puzzle worth mentioning.

One: A number of species in the genus Enterocloster, which sits not far from Ruminococcus gnavus on the phylogenetic tree, are similarly elevated in schizophrenia. Some Enterocloster species also express that aromatic decarboxylase, and these are among the only other common gut bacteria which do. But get this: in that big metabolomics meta-analysis I mentioned, Enterocloster is also tightly correlated with serum levels of N-acyl ethanolamines, endocannabinoid-like compounds that signal via the dorsal root ganglia to downregulate expression of MAO in the brain.5 In people without much tryptamine-production capacity in their gut, this may be a good thing, as it seems to be associated with motivation and athleticism. But since that enzyme is also the brain’s last line of defense against rogue tryptamines and phenethylamines, suppressing it in someone with high background amounts of those compounds sounds like a recipe for a manic or psychotic episode.

Two: After Ruminococcus gnavus, the two microbes most differentially enriched in schizophrenia are Eggerthella and Flavonifractor. These are known for their ability to dehydroxylate and otherwise tear apart catechin and similar flavonoids. That’s all fine when it’s the tannins in your tea that they’re breaking down, but the B ring of catechin looks an awful lot like the aromatic end of dopamine, norepinephrine, and adrenaline—this is why they’re called the catecholamines, after all. In 2020, it was shown that Eggerthella can in fact dehydroxylate dopamine and similar compounds, in the process of using them as electron acceptors. It probably means something that bacteria which can undo the rate-limiting step in the biosynthesis of key neurotransmitters are so heavily enriched in a disease involving severe dysregulation of those neurotransmitter systems, although without more data I’m hesitant to speculate as to what.

Three: In a recent mono-colonization study of mice, Ruminococcus gnavus alone didn’t induce neurological symptoms, but the combination of R. gnavus and liver damage did. Now, if you look at the pathway-level analysis of bacterial genes in the schizophrenic gut, it suggests that alcohol is being produced. And if you look at the gut microbione in NASH (non-alcoholic steatohepatitis, what used to be known as “fatty liver disease”), you'll find that it is not quite as non-alcoholic as the name would have you believe. A number of the same microbes elevated in schizophrenia—including R. gnavus, Enterocloster, and Streptococcus—are also elevated in NASH. They produce ethanol in the gut, and it looks like this is what drives the liver damage in that condition, so it wouldn’t surprise me if that’s a component at play here as well.

Four: In the discussion after my talk, another attendee informed me that her former lab had identified DMT as a metabolite in fecal samples, and that antibiotics could suppress its production—suggesting that it can come from bacteria. There are known enzymes in fungi (and likely in bacteria as well) called HyoA which take tryptophan and repeatedly N-methylate it, resulting in N,N,N-trimethyl tryptophan, or hypaphorine. I am not a good enough chemist to say whether this could plausibly still interact with an aromatic amino acid decarboxylase, but the product of this hypothetical reaction would be trimethyltryptamine, or TMT (“tryptoquat?”)—which sounds like a wild time. I think hypaphorine itself is only possible because the quaternary amine has a carboxyl group nearby to steal charge from, so the more likely possibility to my mind is that tryptamine could also interact with a HyoA-like enzyme, which would stop after two N-methylations. HyoA-like enzymes can be found in the genome of E. coli, and—although that isn’t saying much, because practically any damn thing can be found in the genome of E. coli—it’s worth mentioning that this is another bacterium differentially enriched in schizophrenia, being 16-fold more abundant in the schizophrenic gut than in healthy controls.

Now before I open the floor for questions, I want to point out: these bacteria are present at some level—typically below 1%—in the guts of about half the global population, and they’re linked to a host of other diseases like anxiety, depression, cardiovascular disease, and IBS. It’s the closest thing I've ever seen to an actual, literal demon.

At the beginning of this year, I started a company, to turn ideas like this one into solutions. So far it's just me, but I'm looking for backers and collaborators, so if you'd like to help me wrestle the devil himself and get him in a headlock, find me at the break.

—🖖🏼💩

This is the point in the story where, if I had a PhD, I would publish all this as a paper, maybe do some animal studies, apply for a grant to run a human trial, and then sit on my hands and wait. But I never had the patience for a PhD, and certainly don’t have the patience for all that.

Y’see, this is my mom:

And while I was home for the holidays, I noticed something strange; a pale crescent at the edge of her iris, visible when she glanced down during one of our conversations.

And this freaked me out, because I had seen something like it before—but only once. Back when I was living in the suburbs outside DC, my landlord had a similar ring show up in his eye. I noticed it when I went to visit him in the hospital; I don’t know how long it had been there, he had been unwell for a while…but a few days later, he was dead.

So when I saw the same thing happening to dear old mom, I started frantically googling. As it turns out, it’s a known phenomenon. In the medical literature it’s called arcus senilis. At first, I thought maybe it was a depigmentation, a draining of color—but it’s actually a clouding-over.

Lipid deposits, accumulating in the cornea.

Cholesterol.

Now, like I said, at this point I had already isolated Astrobacillus and done the preliminary characterization work. I was still working for Seed, and for a while I’d been trying to get up the momentum within the company to develop it, but I wasn’t having much luck.

And in that moment, I had this terrible vision of a future where my mom has a stroke while this project is in development hell, and then a year later we find out that Astrobacillus works like gangbusters for reducing stroke risk, and I spend the rest of my life wondering what might have happened if I had moved faster, eventually going mad with regret and becoming a supervillain who calls himself…I don’t know, the Poopler or something.

If I were a perfectly rational, evidence-driven person, seeing that little ring in mom’s eye wouldn’t have changed much. After all, I’ve known for years that she has high cholesterol; statins flare up her arthritis, so she’s been getting by on red wine and warfarin since her valve replacement. And the literature around arcus senilis suggests that it doesn’t actually mean you’re at greater risk for a cardiovascular event than anyone else with unmanaged high cholesterol.

But like all of us, my brain works in emotions and narratives, and there was something so horror-movie about that moment, or maybe fantasy-novel, you know? Young knight errant sallies forth to lift the curse upon the land, slay the jabberwock that’s been killing his countrymen…

…and, having found a piece of the puzzle, returns home to discover that—in his absence—the beast has laid its mark upon his own mother, and the clock is ticking.

Better solve that puzzle quick, boy, the jabberwock grindles.

So we’re trying it!

I was test subject #1, of course—the strain came from my own microbiome (at the risk of sounding, uh, presidential, my lipid panel numbers are fantastic) so that was the logical place to start.

My best friend CJ, who happened to have high cholesterol and no coprostanol in his stool, bravely volunteered to be test subject #2. So far, it seems to be safe and well-tolerated, but the exciting thing is that—even though he last took the probiotic in the first half of May—he still had a coprostanol spot showing up in his poop as of the end of the month. It was a faint band, maybe 10% as dark as the cholesterol one, and another two weeks later it had vanished again—but even this is a pretty solid milestone in my book.

Still, we’ve got a ways to go to reach our goal here, which is engraftment: the bacterium becoming a part of his microbiome the way it’s part of mine, moving him indefinitely from “non-converter” status to “converter”. In theory, this shouldn’t be too hard, given that it’s apparently how things work in every healthy person: at some point early in life, a few cells of Astrobacillus find their way into your gut—likely from your parents’—and take root, reproducing faster than you can poop them out, equipping you with “bonus” genes for the rest of your life…or at least until you get unlucky with an antibiotic.

More promising than “in theory”, it’s already been proven possible in animals. In a 1995 study, researchers put some rabbits on a high cholesterol diet,1 and then fed them some live Eubacterium coprostanoligenes HL (the only coprostanol-producing bacterium currently available in public strain banks) for ten consecutive days. In the month that followed, the control rabbits—which were fed the same diet, but given heat-killed bacteria instead of live—had plasma cholesterol levels averaging 248 mg/dl, while the experimental group’s cholesterol was 183 mg/dl.

That’s huge—the kind of difference that’s not only statistically significant,2 but what we’d call “clinically meaningful”. A drop of 65 mg/dL (AKA 1.68 mmol/L) is similar to the average effect of statins—and studies have shown that lowering your LDL by that much is enough to cut the risk of heart attack by about 60%, and stroke by 17%. Keep in mind that the probiotic was only dosed during the window labeled “Eubact” on the graph above (days 50-60), so clearly engraftment is possible—even when using a strain that originated from another species entirely.

Out of an abundance of caution, we’ve been scaling up the doses pretty slow in our first-in-humans work, and waiting a while between tries. The dose that we gave CJ in early May, which seemed to stick around for a few weeks in his gut, was about equivalent to the daily doses that were used in the rabbit study, scaled to a human’s weight. So this bodes well for the possibility that trying again, giving the same dose given multiple days in a row, could make a more permanent impact on his biology.

It might not be that simple, of course. For instance, CJ’s diet is a little weird, consisting almost entirely of cheeseburgers and coke. In some senses, this is a blessing; it eliminates a huge source of potential variance—there’s a reason lab rats are kept on a homogenous diet of pelleted “chow”. But it might turn out that the strain of Astrobacillus we’re working with needs pectin, or cellobiose, or one of the dozens of other things that are abundant in vegetable matter but not in burgers, to properly engraft: As best we can tell, cholesterol’s role in this creature’s metabolism is as an electron acceptor, the same role that oxygen plays in our metabolism. It might be that you need to eat something bitter once in a while, to stimulate the flow of bile (and thus cholesterol) into the gut. Or it might be that Astrobacillus doesn’t play nice with some organism which is present in his GI tract but absent in mine.

So now that we’ve got a little safety data,3 I’m recruiting for a broader study, which will involve looking at the fecal cholesterol/coprostanol ratio as well as examining the composition of the microbiome by sequencing. This will help us suss out the relationship between a person’s coprostanol-producer status, their microbiome composition, and their serum cholesterol. As we try dosing it into more people, we’ll find out if there are patterns which indicate likelihood of success. Maybe we’ll find that it’s easier to move the needle in people who already have a little coprostanol in their stool—that taking someone from 0% coprostanol/100% cholesterol to 10/90 is actually a lot harder than bringing someone whose baseline is 10/90 to 50/50. Maybe it’ll turn out that everyone who’s a total non-converter is also missing Collinsella or Coprococcus in their guts, suggesting that these people need a combination probiotic. As mentioned last post, though, there’s a pretty strong inverse relationship between the abundance of this bacterium in your gut and your serum cholesterol, so it might be as simple as “more is better”.

In any case, it’s an exciting time, and it’s going to be a busy couple of months. If you’re interested in helping alpha-test this (or just contributing to the dataset) and you haven’t already signed up, you can do so here.

—🖖🏼💩

Imagine being woken in the predawn light, at the break of some prehistoric day, by the sound of a friend saying your name.  You open your eyes, and he’s standing there, staring across the plains into the distance.

“Get up,” he says, not turning his head.

You follow his gaze. After a moment, dim shapes on the horizon resolve: a herd of wild horses.

As you rise to your feet, he turns to you, and now you can see there’s a dangerous glint in his eye. A wild, insane look, with a vision behind it. One of glory, of riding on thundering hooves.

In his hands, he’s clutching a woven rope.
“Come on,” he says. “I’ve got an idea.”

That’s where we are right now. It’s gonna be a big day.

When I wrote those words two and a half years ago, I read it back and thought “ehh, it’s a bit much, isn’t it? Am I being melodramatic?”

But friends, I come before you today a man triumphant: one who set out to wrangle horses, and instead comes home riding a unicorn. A strange and wonderful beast, known from tales in days long past (think, like, 1980), but which has eluded capture for so long that it had started to take on the air of legend. Ladies & gentlemen, it’s my great honor to introduce: the human gut’s native cholesterol-reducer, a microbe much-discussed but never before seen in the flesh…until now. A bacterium that might just save your life.

I’m calling it Astrobacillus, the star-rod.

Cholesterol and Coprostanol: A Quick Review

If you’re a longtime reader, you might remember this piece on cholesterol from back in 2023. For those of you just joining us, the gist is that humans (and all mammals, really) are supposed to have gut bacteria which help us get rid of excess cholesterol.

It’s an important job. On top of the cholesterol in your diet, your liver makes nearly a gram of the stuff each day, to help you absorb nutrients in your food. If it doesn’t leave the body as part of your poop, it can build up in the bloodstream in the form of atherosclerotic plaque. When a chunk of this stuff breaks loose and clogs an important blood vessel, you’ve got a heart attack or a stroke on your hands.

So we rely on these gut bacteria, which enable us to shit out cholesterol by transforming it into a chemical called coprostanol. See, left to its own devices, the body recycles cholesterol very efficiently, secreting it into the gut and then taking it back into the bloodstream along with the fats in your food. The transformation into coprostanol (a reduction reaction) causes it to precipitate, to crystallize in the GI tract, preventing its reabsorption: out of the loop, ‘til it’s out in the poop.

But in the ‘70s and ‘80s, it was discovered that roughly 1 in 5 people are missing this pathway, and have little or no coprostanol in their stool. In the ‘90s, we learned the likely reason why: a standard week-long course of oral antibiotics has about a 10% chance of driving the coprostanol-producers in your microbiome extinct—meaning you basically lose the ability to shit out cholesterol. Hopefully it’s not hard to see how this could become a problem.

This finding is exciting, because it suggests that the root cause of a lot of cases of hypercholesterolemia could be as simple as a lack of coprostanol-producing bacteria. If we can put them back, maybe we can cure high cholesterol.

The only catch? Nobody had ever isolated a coprostanol producer from the human gut before. They’re considered “uncultured”.

In the ‘80s, a handful of coprostanol-producing strains were isolated from the guts of baboons and rodents, but all of these were somehow “lost”, according to the literature. More detail than that is mysteriously hard to find, but these things happen all the time—a freezer fails, or a lab floods, or a scientist dies and nobody picks up the project they were working on.

In 1994, one strain was isolated from a “hog sewage lagoon”,1 and this organism (dubbed Eubacterium coprostanoligenes HL) is—to date—the only confirmed coprostanol-producing bacterium known to man.

But that’s about to change.

Materials and Methods

Isolating this guy wasn’t exactly easy, but it also didn’t require any great leaps of insight; just diligent research (shoutout to Alexandra Elbakyan at Sci-hub), an anaerobic lab (shoutout to Raja Dhir at Seed), and a willingness to do some things that, to your average person, probably look insane.

See, when I wrote that first piece on cholesterol and coprostanol, I had already found the papers from the ‘70s where they described isolating the first coprostanol-producing bacteria in liquid culture—a painstaking process of serial dilutions and antibiotic treatments, because the organism refused to grow on solid agar in a petri dish.2 What I hadn’t yet found when I wrote that post was an article by Brinkley, Gottesman, & Mott (1980)—a paper from just a few years later, where they tell you exactly how to make an agar that coprostanol-producing bacteria will grow on.

The secret ingredient is “way more cholesterol than seems reasonable”…and a healthy helping of mammalian brain.

But once the brain has been freeze-dried and papain-treated, and the special agar has been made, nabbing one of these bacteria was as simple as taking a shit, diluting it by a factor of ten million or so, spreading a little of that water on the petri dish, and—a week later—looking for the “fibrous” morphology described in the paper.

A Little TLC

The fact that this bacterium was a visual match to the cholesterol-reducers described in the literature was a good start. Sequencing its 16S gene—sort of the bacterial “barcode”—revealed that it wasn’t very closely related to any named organism, but one of the closest was Eubacterium coprostanoligenes. This is even better, but it’s still what a lawyer would call circumstantial evidence. I wanted proof-positive.

Unfortunately, I’m not a very good chemist, and I didn’t have ready access to the kind of equipment typically used to interrogate an organism’s molecular makeup. The Swiss Institute for Allergy and Asthma research—where I was when I first isolated one of these guys—has plenty of mass spec rigs, but they’re all set up for proteomics, i.e. looking at things thousands of times the mass of individual molecules like cholesterol and coprostanol. Lucky for me, back in 1987 some folks at the FDA published a low-tech method that turned out to be just what I needed: cheap, easy, and hard to fuck up.

See, “is this a piece of poop or not?” is a million-dollar question when you find a weird chunk in your corn flakes, so the gold standard way to find out is to check for the presence of coprostanol, since it’s abundant in mammalian feces and not really found anyplace else. The method devised at the FDA relies on thin-layer chromatography (TLC), which is about as simple as analytical chemistry gets; if you ever did the coffee-filters-and-marker experiment in school, you’ve got the gist.3 You put your sample on a plate of powdered silica gel (of DO NOT EAT fame), dip the end into a solvent, and let capillary action pull that liquid up against gravity—carrying the chemicals in your sample along with it. Because molecules move up the plate different rates depending on their structures and properties, they show up as distinct spots.

To complete the picture, it helps to have a little context for these dots: that top lane is my own poop sample, and the middle two are my friend Cole’s, from samples taken on different days. I’ve got no cholesterol in my poop, Cole has no coprostanol.

These are two ends of a spectrum; I’ve run this assay on maybe two dozen people so far, and most people have both spots—some people are 90:10, others are 40:60—but my results so far basically agree with the numbers given in the literature: about one in five people are like Cole, in that their microbiome doesn’t seem to produce coprostanol.

But the bottom lane in that image is what you get when you inoculate one of those star-shaped colonies into cholesterol-brain medium and incubate it a few days. The sterile medium starts out looking like those middle lanes—all cholesterol—but by a week later, the coprostanol spot stands out, clear as day.

Gene sequencing the bug reveals that it’s the first isolate of its genus—one of those bacteria that’s currently just denoted by an alphanumeric string wherever it crops up in metagenomics datasets. This is exciting for a couple of reasons, one being that—as the first person to culture one in recent memory—I get to name it. I like the sound of Astrobacillus, for its star shape; we’ll see if that flies with the nomenclature people. (I’m pretty sure it will, but they’ve got a lot of rules and my Latin isn’t great.)

More exciting, though, is having the full genome of this organism, because it acts as a sort of Rosetta stone—turning otherwise-meaningless data into something really useful.

A Brief Foray Into the Weeds

See, over the years, a number of research groups have done coupled metagenomics-metabolomics studies, where they look at the bacterial genes in people’s guts as well as the chemicals in their blood, and search for patterns: for every bacterium X and metabolite Y, you check: is there a correlation between the amounts of these two things in people’s bodies?

And the answer is “yes!” for a lot of X<>Y pairs. Too many, in fact, to even know what to do with most of the information. But recently, someone did a meta-analysis of ten of these studies, and for good measure they threw out any signals that weren’t consistent across at least three datasets. That still left nearly 30,000 microbe<>metabolite pairs, and the researchers were kind enough to publish all that information as a big beautiful excel sheet in their supplemental materials for anyone depraved enough to comb through it. Depending on how much you know about a given metabolite and a given microbe, the info contained in datasets like this can range from “abstruse” to “intriguing yet inscrutable” to “wildly useful”.

For a simple test case, we can look at butyrate: it’s a short-chain fatty acid that’s abundant in the bloodstream, which comes pretty much exclusively from your gut bacteria. You can read about SCFAs anywhere, so for now suffice to say it plays a number of important roles in keeping the body running smoothly.

Filtering the table to look exclusively at bacteria that are associated with butyrate, and then sorting by effect size, you can see in the screencap above that bugs like Agathobacter, Roseburia, and Faecalibacterium are the genera that best correlate with butyrate levels in the bloodstream. And this is great—it’s a “pass” on the sanity check: all three of these are known powerhouses for butyrate production, and pretty universally associated with “healthy” status in studies of the gut microbiome.

But say we sort the table the other way, and look at inverse associations. What does it mean that Ruthenibacterium is the genus that best predicts low butyrate levels?

Ruthenibacterium isn’t nearly as well-characterized. Does it eat butyrate? Prevent its formation? Or is it an innocent bystander that simply thrives in a low-butyrate microbiome? It’s hard to say, without doing a lot of wet-lab work—and that goes double for things like “UBA5446” on that table: these are the uncultured bacteria. We don’t even know what they eat, or what they turn their food into, and there’s no way to find out for sure, until someone manages to nab one in culture so it can be poked and prodded. If it turns out that the bacterium can grow using pure butyrate as a food source, hey, now we have a pretty good idea as to why it’s associated with low levels of that molecule in the bloodstream.

So when I genome-sequenced this coprostanol-producing bacterium, I found out that it shows up as “CAG-177” in tables like this one. And if we do a similar filter-and-sort trick, to see what chemicals it’s associated with in the bloodstream—

Turns out the abundance of this bacterium in your gut is inversely associated with serum cholesterol levels! You can see that it’s also correlated with a number of other compounds, but of all the chemicals it’s linked to, it’s most strongly tied to cholesterol. Not only that, it turns out this is the strongest association between serum cholesterol levels and any gut bacterium in the table.

In other words: the more of this bacterium you have in your gut, the less cholesterol you tend to have in your bloodstream. Before we knew that CAG-177 was a human-gut-native coprostanol-producer, this was one of those “intriguing yet inscrutable” findings. Now, it becomes solid evidence that coprostanol and the bacteria that produce it are a really important factor in cholesterol homeostasis, and maybe even heart disease risk.

There are still ways this might not be causal in the way that I’m hoping; there’s a world where Astrobacillus only thrives in the gut when your HDL particles are doing a really good job of scrubbing cholesterol out of the bloodstream and depositing it in the GI tract or something. Curiously enough, if you filter-sort the cholesterol x bacteria table the other way—to look at bugs associated with high cholesterol—you’ll find Ruminococcus gnavus at the top of the list. This is a known baddie that I’ll be talking about more in an upcoming post, so I won’t dwell too much on it here—but it’s something I’m chewing on. Maybe one of these bacteria eats the other, or inhibits its growth. Maybe R. gnavus modifies bile acids in a way that prevents your liver from excreting cholesterol into the gut, and this is bad for Astrobacillus abundance. Microbiome science is a kind of ecology, and ecology is always complicated.

Even so, the fact that we’ve found a bacterium that…

a) ought to help lower cholesterol levels, based on everything we know about its molecular biology, AND

b) is associated with low cholesterol levels in multiple actual human cohorts

…is exciting.

Reflections

I struggled for a while to find the right ecological analogy for these guys, but I think that in a lot of respects they’re like the giant panda. Not the fittest, in a lot of senses; slow growing, slow moving, and their milk is so nutrient-poor that—if a mother has twins—she has to pick her favorite and let the other one starve. When all you eat is bamboo, which is about on par with driftwood in terms of nutritional value, it takes a long time to save up enough calories to afford a baby. The flipside is that, when all you eat is bamboo, you never need to worry about where your next meal is coming from. It’s such shitty food that nobody else even bothers with it, so your entire native range is an all-you-can-eat buffet, literally within arm’s reach 24/74. Having exclusive access to an extremely abundant food source goes a long way in making up for whatever you might lack in other domains.

Cholesterol is like bamboo in that, in the microscopic jungle of the gut, it’s extremely abundant. And even more important, a steady supply of it is pretty much guaranteed. Bugs that specialize in the degradation of starch or other complex carbs are subject to the whims of their host’s diet, meaning their population can suffer a serious blow if, say, you go a week without eating a piece of fruit. But regardless of what you’re eating, as long as you’ve got a functioning liver, you’ll have cholesterol in your gut.

There’s lots more to get into here, including some exciting human-subjects-research results to share, but this post is already long enough—stay tuned for the next installment!

—🖖🏼💩

EDIT: Part 2 of this piece is up!

Lind was a shipboard physician in the British navy, and had eight sailors with scurvy on his hands. He gave two of them vinegar, another two dilute sulfuric acid, two more got seawater, and the remaining two got citrus fruits. When the citrus group made a spectacular recovery, Lind knew he was on to something—so a few years later, when the famous captain James Cook was setting out on a voyage, Lind sent him off with a crate of lemon juice for his sailors, boiled and bottled to prevent spoilage on the long journey.

When Cook returned, he reported that Lind’s juice was bunk—no difference in scurvy rates between sailors who drank it and those who didn’t. Vitamin C, we now know, degrades when you boil it.

And so it was that the first-ever RCT was immediately followed by the first-ever failure to replicate exciting preliminary results, thanks to sloppy design when scaling up an RCT. It wasn’t ‘til 1779—more than thirty years after Lind’s original trial—that someone thought to try preserving the lemon juice with alcohol instead, and found that this maintained its scurvy-preventing magic.

I think a lot about that thirty-year gap. That lull, where the key discovery had been made—the cure for a horrible disease was this close to existing—but the implementation needed a little workshopping. How many men died grisly deaths, old scars turning into gaping wounds as their skin literally unraveled at a molecular level, all because Lind and Cook’s ill-fated test threw everyone off the scent? How many lives were lost for want of a technology that was, essentially, a pre-mixed margarita?

How many things like that are out there, right now? How many puzzles with life-and-death stakes could be solved, if someone would just pick up the pieces we have lying around and tinker with them a little?

In the last decade or so, we’ve learned a lot about the gut microbiome. So far, all signs indicate that, if we can learn to manipulate it with enough finesse, it has the potential to revolutionize human health. When we gene-sequence the bacteria in a person’s poop, we find that—in practically every poorly-understood health condition—some common species are consistently missing, or others are present in excess.

So what do we make of the finding that lower abundance of Eisenbergiella in the gut is associated with worse seasonal allergies? Eisenbergiella is one of those bacteria we’ve had with us since the caveman days; maybe it’s important. Maybe it produces a natural antihistamine or something. Why is Fusicatenibacter saccharivorans depleted in a smattering of autoimmune diseases with weird names, like spondyloarthritis? Why is Ruminococcus gnavus some 30x more abundant in the guts of people with schizophrenia than in controls? (that one, I have ideas about; stay tuned.)

The skeptic’s refrain goes “Correlation is not necessarily causation”—but it’s a stone’s throw from there to nihilism; the sentiment that nothing ever happens and the world is not for us to understand. It’s possible—likely, even—that some of these relationships are causal: that a change to the microbiome is the driver of disease, and changing it back could set things right.

This is a hypothesis that, in theory, doesn’t have to be that hard to test. You find some people with allergies, you feed them some Eisenbergiella, you see if it helps. If not, hey, see if you can find some people with spondyloarthritis.

So why hasn’t it been done? Why are the drugstore shelves lined with dozens of probiotic brands, all of which contain the same old yogurt starter? Why are there a dozen companies selling microbiome tests that’ll helpfully inform you that you’ve got no Coprococcus in your gut—but zero companies that will actually sell you a Coprococcus probiotic?

Why, if all the research is so promising, does it seem like nobody’s actually doing anything with it?

Well, for one thing: looking at the microbiome and doing something about it are two entirely different disciplines. The former, the computational side, is what’s boomed since ~2014, when gene sequencing got cheap enough that people could afford to poop into an Illumina for kicks. But actually growing the bacteria so you can feed them to people hasn’t gotten any easier since about the ‘80s. See, most of the “good guys” in your microbiome are strict anaerobes, which means it takes specialized equipment and an uncommon skillset to culture them. It’s not hard, per se, but it’s not something you can easily pick up without hands-on training by someone who knows what they’re doing.

So it’s a small field. In the US, there’s a handful of companies and a few dozen academic labs that have the equipment and know-how to culture the strict anaerobes of the human gut—the Faecalibacterium, the Bacteroides, etc. Sometimes, those university labs will get as far as running a rodent study on an interesting strain, but making the jump from there to a human trial is a massive undertaking; on top of the science itself, the admin and oversight required is a full-time job, so it’s not something you typically do without first winning a grant to fund it.

The private sector could have less red tape: A corporation isn’t beholden to IRBs and funding bodies the same way academia is—and since probiotics are regulated as a food in the US, developing a new strain could be as simple as bringing a new kind of cheese to market. But so far, most companies working on turning microbes into marketable products have tried to develop them as drugs. And the admin and oversight involved in that? That’s an entire company’s worth of full-time jobs. (Oh, and at the first hint of an underwhelming clinical trial, the whole thing implodes spectacularly and everyone loses those jobs.) “Ingredients” companies like Biose and Sacco—the guys that make the Bifido powder that goes into those zillion different whitelabeled brands—are starting to get their gears turning on the anaerobes problem, but they don’t typically run a large study until they’ve got a shelf-stable, market-ready product…which means putting in months or years of upfront R&D work before each attempt on goal.

Long story short: observational research is about ten times cheaper and easier than interventional research, so it progresses about ten times faster. As a result, there’s a huge gap: a lot of promising hypotheses that nobody’s gotten around to properly investigating yet, simply because there hasn’t been time—especially when scientists capable of doing so are few and far between.

But I am one such scientist, and so I’m starting a company to turn all this research into something real: to put these correlations to the test, as directly and efficiently as possible, find out which ones really are causal, and get those bacteria on the market and into the hands of people who need them.

My goal is to create something less like probiotics as they exist today and more like gene therapy. These will be precision products, designed to fulfill a specific ecological and biomolecular niche, rather than something vague like “support gut health”. They will not be freeze-dried powder in a pill. They will, with any luck, be one-and-done solutions: you take enough that the bacteria set up shop in your intestines, eating your food and reproducing, providing health benefits for months or years to come. This goal—engraftment, as it’s known—is considered by many to be the “holy grail” of microbiome science, but it’s not too high a bar; after all, this is how all the bugs already in your microbiome do it. (Really, it’s outrageous that the current probiotics paradigm has people renting their gut bacteria. In this age of everything-as-a-service, of “You’ll own nothing and like it,” don’t we deserve to at least own our living selves?)

I’m calling the company Constellation.

I’m raising money—if you’re a high-net-worth reader and you'd like to back me, reach out—but I’m not waiting for a round to close to get started. I’ve already got the lab up and running, and we’re in human testing on our first product. Naturally, I tried it first; it’s a strain isolated from my own microbiome, so that was the logical place to start the safety tests. But now we’re on to figuring out how much it takes to get engraftment in someone who’s missing it. We’ll talk more about the specifics of this project in the next piece, but if you want a sneak preview, check out Constellation Bio’s website and reread my old post High Cholesterol? There’s a Bug for That.

But I think this is important to highlight: we are skipping the in vitro cell culture models, we are skipping animal tox, we are skipping freeze-drying and the dozen other in-between steps that ostensibly move the ball downfield but, in the end, mostly serve to get grants, keep postdocs busy, and put off the day when you risk finding out that your hypothesis was wrong.

We are doing it quick and dirty, going straight into humans, taking shots on every goal in sight. We are doing it with rigor and urgency; with alacrity.

We are doing this because, while there are risks in taking any drastic action, risk is only the possibility of harm, and it amounts to nothing when weighed against the certainty of harm we face if we don’t act. How many people will have a stroke between now and when this cholesterol bacterium is ready?

Too many.

I’ve written before about the difference between opinions and convictions: An opinion is something that’s consistent with your worldview, that you’d say you believe if asked—but a conviction is something that you live by, that you really act like you believe. And while I see a lot of people talk about the microbiome holding the key to the future of health, I don’t see anyone acting with the urgency of someone who truly believes: We might be able to cure high cholesterol, the thing that is statistically the most likely to kill you and your loved ones. We might be able to rid people of depression, yank it out by the root, save them from the suffering that drives people to suicide. To lift the rock of anxiety off a friend’s chest, put the freshness back into the breeze on their skin. Against such urgency, multiplied by millions of souls struggling under those burdens—is any action too drastic?

We know, from the story of scurvy and a dozen other stories like it, that dedicated scientific effort can vanquish horrible diseases. A skeptic might say the vitamin deficiencies were low-hanging fruit; that it’s easy when the cure is “eat an orange once in a while”. But a lot of our gut bacteria are like vitamins, in that the body expects to have them around, and has for so long that it’s not meant to function without them.

So this is my conviction: these bacteria are our birthright. They’re pieces of us. And it’s only through the vicissitudes of modern life—antibiotics, atomization, the industrialization of food—that we’ve begun to find ourselves without them.

What miracles are within reach, if we can manage to give people back the bits of themselves that they’ve lost?

We’re going to find out. Let’s get to work.

In most cases, this gets a grimace, coupled with an “ehhhhh” noise, and the kind of hand-wiggling motion which—at face value—means “so-so”, but which in this context indicates “I have next to no idea how this thing on my shoulders works.”

So it’s usually with a relieved smile that they’ll answer my next question: “You know about dopamine?”

Oh, easy. Everyone knows about dopamine.

“And serotonin? You’re familiar with serotonin?”

More enthusiastic nodding—of course. We’re two for two!1

“Good!” I tell people. “You’ve already got all the background you need.”

Why and How

Whether or not they get all the details, nearly everyone is on solid footing with the understanding that these neurotransmitters are important for mood, memory, and the ability to think. Maybe you’re aware that, when people talk about SSRIs being the most widely used kind of antidepressant, one of the “S”-es stands for “serotonin”. If you’re a real overachiever (or an intellectually curious degenerate), you’ll be aware that drugs like cocaine and amphetamine work by activating the serotonin, dopamine, and norepinephrine systems.

Knowing that these drugs tend to impart energy, enthusiasm, and the feeling that everything is cool and froody—and that they do this by mimicking those three neurotransmitters—it’s all too easy to understand why it’s catastrophic if something goes wrong with your body’s ability to produce them. This is the “neurotransmitter imbalance” theory of depression, and—even though we finally got direct evidence recently that depressed people’s brains really do release less serotonin than healthy people’s2—this explanation still kicks the can down the road: It’s a how, not a why. If a neurotransmitter imbalance causes depression, what causes the neurotransmitter imbalance?

To get the obvious caveats out of the way: a zillion different things—from heavy metal poisoning to genetic mutations to infections—can mess with neurotransmitter biosynthesis. Talking about depression as if it’s a single, explicitly biological condition necessarily paves over a lot of nuance: Odds are that maybe 40% of people with the diagnosis “major depressive disorder” have one specific thing wrong with them, another 30% have some completely different issue which happens to result in similar symptoms, another 10% have some third thing, and so on. Add in psychological and social factors, and the picture gets even worse. The same is true of schizophrenia, bipolar, autism, and a host of other psychiatric, neurodevelopmental, and personality disorders. This is one big reason why none of these diseases are well-understood at the root-cause level yet.

But another reason is that practically everyone trying to understand these diseases over the last century has been starting from the premise that—if we want to learn about them—we should be studying the brain. That work has turned up a number of leads and insights: differences in the size or connectivity of one lobe or another, neuron growth factors like BDNF being higher in healthy controls, and the famous neurotransmitter imbalance…but the closest we’ve come to a why are things like “oxidative stress” and “inflammation”, both of which leave a lot to be desired by way of actionable solutions. (“More exercise, more sleep, and uhhh, drink green tea?”)

In recent years, though, we’ve begun to see solid evidence that—although the how certainly unfolds above the shoulders—the why might lie a little lower down, in the gut microbiome.

It’s an intuitively pleasing idea for a lot of reasons, from truisms about “gut feelings” to the well-established associations between diet and mental health—but first and foremost is that the microbiome is the one place in human biology that we haven’t really looked yet. The genome has been checked pretty thoroughly, and for the most part ruled out: there’s no reliable way to tell whether someone has depression or schizophrenia just by looking at their genome. This fact has been driving scientists batty for decades, because most of these diseases ARE very clearly heritable: if you know whether a person’s parents or siblings have a mental illness, you’ll have about ten times better luck predicting their mental health status than the best genomic analysis can do.3

This discrepancy has led some scientists to theorize that there must be thousands of different genes involved, each of which influence the risk of mental illness a tiny bit—making the impact of any one hard to suss out. But over the last ten or twenty years, a much simpler explanation has emerged, with the discovery that the gut microbiome is not only highly heritable, but wildly variable across the population—and even within a given person over the course of time.

That heritability helps explain the genome discrepancy, and—more importantly—it makes it seem a lot more plausible that the microbiome could be necessary for key biological processes. Before we understood that core components of the microbiome are passed down as reliably as any of our own genes, it seemed unlikely that we would rely on gut bacteria for critical things like “staying sane” or “wanting to live”.

But sure enough, as the technology has gotten good enough for us to just shit directly into a gene sequencer and see what’s going on in there, we’ve found that the gut microbiome is out of whack in a number of psychiatric diseases—and that here, too, you can do much better at predicting whether someone is mentally healthy vs. ill than you can by looking at their genome. This is extremely exciting for two reasons:

First off, it offers the prospect of an ultimate, simple, satisfactory why. A disruption to your interior ecology is to blame—some bacterium that’s there where it shouldn’t be, or missing where it was present in every one of your ancestors. We’ve all taken antibiotics at some point in our lives, which is a little like playing Russian roulette with the gut’s ecosystem: Most of the time, nothing bad happens—but there’s always a nonzero chance that you’ll drive an important species extinct. If it turns out that species was essential for helping you make dopamine or serotonin, maybe you end up with depression.

That’s a simplified example, but it brings us to the second reason this is exciting: the microbiome is also infinitely easier to modify than the genome. If we find a bacterium that’s absent in some people with depression, but present in everyone healthy, replacing it might just fix things for those people.

So here we find ourselves “kicking the can” from the other direction: with a plausible why locked in, we can look for how something wrong in the microbiome might lead to a neurotransmitter imbalance.

The better part of a decade ago, I went out looking for that how, and I found one. A direct route from the microbiome to the monoamine neurotransmitters—serotonin, dopamine, and norepinephrine. No hand-waving, no unknown mechanisms: a clear hypothesis, and—if it turns out to be true—a simple solution. This is the idea that made me quit my job as a science writer, file a patent, move to Boston, and dive headlong into shit, so to speak. An idea so compelling in the scope of its implications that, even if there were only a one-in-ten-thousand chance of it being correct, it would be worth reorienting my entire life to try and find out.

This is the Vitamin Q Hypothesis.

What Could Go Wrong?

The catalyst for this grand idea, which hit me in about 2016, was a scientific article from 2011. The moment I read the title, I was thunderstruck—it was suddenly right there, clear as day—and by the time I’d finished reading the abstract, it took everything I had not to jump out of my seat in the café and shout eureka. We’ll get to that article and its magic title in a little bit, but we’ve got some ground to cover first before it’ll truly make sense.

So let’s start with the basics: serotonin and dopamine. Chemical cousins, they look somewhat similar, but they’re independent of each other: Each lands on its own unique set of receptors, which do different things in different parts of the brain. Serotonin doesn’t activate the dopamine receptor, and vice versa. Nevertheless, drugs that activate one often activate the other, thanks to their structural similarity.

That resemblance is no coincidence: although each is produced in the body from a different amino acid (serotonin comes from tryptophan, and dopamine comes from tyrosine), those parent amino acids are both produced by a set of molecular machinery referred to as the shikimate pathway. Humans don’t have this machinery, so tryptophan, tyrosine, and the related amino acid phenylalanine, are all considered essential amino acids: our bodies can’t make them from scratch, so we have to eat them (although the microbiome probably kicks in some—see the Thousand Secret Ways).

OH!

Now, even if you’ve never taken a chemistry course, maybe you can already see the important thing here: that these two critical neurotransmitters, dopamine and serotonin, are produced by a perfectly parallel process. It’s the same set of two simple transformations, carried out on two different starting materials.

The addition of a hydroxyl (OH) group turns tryptophan into 5-hydroxy tryptophan, or 5-HTP. This is the immediate precursor to serotonin.4 Installing an OH group in a similar spot on tyrosine turns it into L-DOPA, which is the precursor to dopamine. Sharp-eyed readers might have noticed that the second step in the diagram above—the chemical reaction that turns 5-HTP into serotonin—is also basically the same as the one that turns L-DOPA into dopamine: the removal of a CO2 from the “teeth” end of that key—a decarboxylation.

We’re not going to focus on that, though, because it’s the first step—the hydroxylation—that’s the important one here. This is the hard part for your body, the rate-limiting step in the synthesis of serotonin and dopamine in your brain: the cap on how much of these chemicals you can produce. And because you need serotonin in order to produce melatonin (and dopamine to produce norepinephrine and adrenaline), that chokepoint has the potential to impact a cascade of processes from mood and cognition to sleep.

That same hydroxylation reaction, it’s worth noting, is also the key step in preventing the amino acid phenylalanine from being neurotoxic: Because it looks a lot like tyrosine and tryptophan, phenylalanine can block up the transporters that are supposed to carry tryptophan and tyrosine from the bloodstream into the brain—leaving your neurons starved for serotonin and dopamine. Phenylalanine is abundant in most high-protein foods, but in a healthy body, an enzyme in the liver hydroxylates most of it into tyrosine before letting it into circulation. If that enzyme doesn’t work properly (some people are born with a “broken” version, thanks to a genetic mutation), you get a disorder called phenylketonuria, or PKU—where you have to religiously avoid phenylalanine, or else you end up with social and intellectual disabilities, a weirdly pale complexion (tyrosine is also the precursor to melanin), and a “musty” odor radiating off you.

All this to say: if we want to understand why some people seem to have a hard time sleeping at night, waking up in the morning, focusing in class, or just generally feeling like a human being ought to, it’s worth taking a close look at that rate-limiting step—at the little molecular machines that grab a tryptophan, or a tyrosine, or a phenylalanine, and stick an OH onto it.

These are the aromatic amino acid hydroxylase enzymes (or “AAAH”, if you like).5 There are three of them; a dedicated one for each of those three amino acids. Tryptophan hydroxylase turns tryptophan into 5-HTP, phenylalanine hydroxylase turns phenylalanine into tyrosine, and tyrosine hydroxylase turns tyrosine into L-DOPA.

Got it? Good.

Batteries

These three enzymes are pretty similar to one another, both in structure and in function. They’re almost identical, in fact—which means that all three run on the same fuel, a molecule called tetrahydrobiopterin.6 We’ll call it BH4, for short.

If these three AAAH enzymes are molecular machines for making neurotransmitters, BH4 is the battery that powers them: it’s a cofactor for the enzymes, meaning they can’t work unless they’ve got a molecule of BH4 attached in the right spot. Right off the bat, this seems like an important molecule, doesn’t it? Sitting at the corner of serotonin street and dopamine drive? Where do we get more of it?

Truth be told, you don’t want more biopterin. Your body can make BH4 from scratch, but making it isn’t the hard part.

Just like a battery, BH4’s charge runs down when it gets used: Every time your body converts a molecule of phenylalanine to tyrosine, or tyrosine to L-DOPA, or tryptophan into 5-HTP, a molecule of tetrahydrobiopterin gets “burnt up” in the process—it oxidizes to dihydrobiopterin, or BH2.

BH2 is like a dead battery in the junk drawer. It looks enough like BH4 that it can still slot into the right spot on the AAAH enzymes, but it just sits there—no juice, no serotonin. Eventually, that molecule of BH2 will drift off, and then there’s the chance to try again—but there’s nothing stopping another molecule of BH2 (or even the same one!) from drifting back in. I guess your body, in this analogy, is the kind of degenerate who will take a battery out of the drawer, try it in the remote, and—if it doesn’t work—throw it back in the drawer before grabbing another to try.

The important thing to understand about this dynamic is: how well the AAAH enzymes do their jobs is less about “do you have enough batteries in the drawer” and more about “what percent of these batteries have any charge?”

Now, your body has a good reason for keeping those dead batteries around rather than just throwing them out: you have enzymes that recharge them, reducing BH2 back to BH4 and making it usable again. Recharging the oxidized cofactor, rather than making a new BH4 from scratch every time you need one, is more biologically cost-efficient (imagine how much you would spend on batteries if your smartphone ran on single-use AA’s) but it’s also a logistical necessity, thanks to the “junk drawer” nature of the cell, and the semi-random way the AAAH enzymes bind their biopterin cofactor: the more batteries you have in the drawer, the harder it gets to improve your odds of getting a good one by adding in charged ones.

So there’s a cycle here, between the AAAH enzymes that oxidize BH4 to BH2—churning out tasty neurotransmitters in the process—and dihydropteridine reductase, the enzyme that reduces the BH2 back into usable BH4. At any given time, you have both forms of biopterin in a cell, at some ratio that varies depending on the relative activity of the enzymes that use up BH4 and those that regenerate it.

There’s a third factor here though, because BH4 is also very susceptible to nonenzymatic oxidation—meaning any free radicals floating around the inside of your cell can also oxidize it to BH2. This is part of how certain heavy metals (lead, in particular) mess with cognitive and psychiatric health: The concentrations of these all these factors in your tissues are such that anything which pushes the ratio of BH4/BH2 towards BH2 will slow down production of 5-HTP and L-DOPA, while anything which pushes it towards BH4 will speed it up.

And the same way that a paperclip in the junk drawer can short out a battery and drain its charge, reactive metals like lead can oxidize your BH4 to BH2 without doing anything useful—making it harder for the reductase enzyme to keep up. Lead isn’t unique there, though; iron can do the same thing if it’s not contained in a structure like heme.

And the same goes for any reactive oxygen species, like hydrogen peroxide generated by the immune system—or even the everyday byproducts of cellular metabolism that can build up if a cell is overtaxed.

When everything’s working properly, free-radicals are kept in check by antioxidant enzymes like thioredoxin and molecules like vitamin C or glutathione, and oxidized BH2 is enzymatically recycled into BH4, leaving you a steady supply. But the complexity of these dynamics is why, at the beginning of this section, I said “you don’t want more biopterin”: taking supplemental BH4 doesn’t really help increase levels of BH4 in your neurons, because most of it gets oxidized before it’s transported into cells—meaning you end up with a whole lot of excess BH2, which actually inhibits the enzymes we want to encourage. The hard part isn’t making enough BH4: it’s keeping it in the right oxidation state.

Fragile By Design

All this seems like a bit of a design flaw, no? Wouldn’t it be better if our mood and cognition didn’t depend on a molecule that’s extremely susceptible to random degradation?

Well…no. Think about the dynamic this sets up: it makes mood into a sensitive barometer for oxidative stress. It means that, if your cells aren’t happy at a molecular level, you’re not happy at the macroscopic level.

You can imagine the benefit this would have in an early-human society. Say your tribe moves into a new area; there’s lush vegetation and abundant fruit, plenty of wild game, and a freshwater spring. Maybe you wonder: Why has nobody else settled here? Did we just luck out?

But suppose the answer is No, because it turns out that the soil here is very high in lead—so the fruit, the game, and the water are full of it. In a world where serotonin and dopamine biosynthesis are independent of oxidative stress, you might not notice anything wrong until a year or so later, when the stillbirths start, and the kids who survive come out with strange defects. By that point, your whole tribe has accumulated enough of this cellular poison to increase their risk of cancer, infertility, etc. Even if you do move away, your elders will die early, taking their wisdom with them.

But in our world, where the serotonin and dopamine synthesis pathway is so exquisitely sensitive, the signs that something is wrong start to show almost immediately: The elders can’t sleep. Husbands bicker with their wives. Children cry inconsolably over nothing. It doesn’t take long for your people to conclude: this land is cursed, we must go. In this example, it’s clear that depression is not a maladaptive trait—it’s your body’s way of letting you know that something subtle is disastrously wrong, and you need to make a change.

But when the thing that’s wrong is internal, rather than external, it’s not as simple as moving to a different neck of the woods.

A Few Protons Makes All the Difference

Now, I am not the first person to recognize that biopterin issues are a promising candidate to explain the symptoms of certain mental illnesses. It doesn’t take a genius to look one or two molecules upstream of serotonin and dopamine and go “hey, what about this thing?”

But if you search the literature for “tetrahydrobiopterin+depression” or “tetrahydrobiopterin+schizophrenia”, you’ll find something curious: a preponderance of studies reporting elevated levels of biopterins in blood and urine samples from patients with these diseases. This, it seems, is enough to convince most researchers that this path is a dead-end. If someone’s already tested that, and the result was actually the opposite of what you expected? The issue must lie elsewhere. But if you read the materials and methods sections of these papers, you’ll notice a pattern: none of these studies are actually measuring BH4.

See, as we discussed earlier, there are multiple forms of biopterin in the body, and BH4 is sensitive to spontaneous oxidation: it doesn’t take much for it to lose a few protons and become BH2, or BH0. Oxygen, light, and time are enough to do it, at room temperature—so if you want to actually measure someone’s urinary BH4 levels accurately, you have to have them piss directly into a cup pre-loaded with three different kinds of antioxidants, in the dark, and then put the sample directly into a deep-freeze within a few minutes, until it’s ready for analysis.

This is a pain in the ass, so practically nobody does it. Instead, the protocol used in all of these papers goes something like:

-Collect urine sample as normal

-Treat it with iodine to deliberately oxidize all the biopterins to BH0

-Analyze to quantify biopterin

-Consult the literature to find the proportion that’s normally present as BH4 vs. BH2 vs. BH0.

Put simply: most people trying to investigate this have traded rigor for convenience, precisely at the most important point, because they aren’t giving the complexity of the system due consideration. You can see the evidence of this in the fact that—in the one or two studies that have taken the care to do it properly—you find higher total biopterins in the urine, but significantly lower levels of BH4 in depressed people than in healthy controls.

The same is true in schizophrenia: the BH4/BH2 ratio is all out of whack. This might come as a surprise if you’re only familiar with the disease from popular depictions—which tend to focus on the so-called positive symptoms: delusions, hallucinations, paranoia, etc. But in the vast majority of patients, these are accompanied by a suite of negative symptoms, which look a lot like a severe case of depression: emotional flatness, the inability to find pleasure in things, executive dysfunction, social withdrawal, and so on.

To make matters worse, the most effective drugs for managing schizophrenia’s positive symptoms are dopamine antagonists, which—as you might expect—make people feel like shit, and make the negative symptoms worse. This is why a lot of neuroscientists talk about the “dopamine paradox” of schizophrenia: the negative symptoms are characteristic of a deficit of dopamine and serotonin, but the positive symptoms can be quelled by drugs that prevent dopamine from doing its job.

The most plausible resolution to this paradox involves another molecule, one called kynurenic acid.7 This is one of the things that tryptophan can turn into, in your brain, if it doesn’t turn into 5-HTP first.

Kynurenic acid levels are higher in people with schizophrenia, which is particularly interesting because the molecule acts as an antagonist at the NMDA receptor—putting it in the same pharmacological class as drugs like PCP or ketamine, which are known to cause delusions, hallucinations, and the sensation that you’re really just a spinal cord piloting a meat suit. And because NMDA antagonists can cause dopamine-producing neurons in some portions of the brain to fire frantically, this is one of the best-accepted explanations for the dopamine paradox: some cellular defect impairs the production of both dopamine and serotonin, leaving your brain’s tryptophan with nowhere to go, except down the path that leads to kynurenic acid. Since kynurenic acid acts partly by causing hyperactivity in certain kinds of dopamine neurons, dopamine antagonists would be useful in preventing its effects.

That Magic Title

Okay, we’re ready. If you’ve been paying attention, you now have all the knowledge about this that I had when I punched a few key words into Google Scholar and had my life changed forever.

It was a simple query; something like “gut bacteria tetrahydrobiopterin”, and right there on the first page of results was this paper from 2011:

Queuosine Deficiency in Eukaryotes Compromises Tyrosine Production via Increased Oxidation of Tetrahydrobiopterin.

Now remember, the body ordinarily converts most of the phenylalanine you eat into tyrosine, with the help of one of the AAAH enzymes—which run on BH4. So this paper is reporting accelerated conversion of BH4→BH2, severe enough that it prevents the AAAH enzymes from working properly.

But what—I thought to myself as I clicked the link—the fuck is queuosine? How is that even pronounced?

Queuosine (“cue-uh-seen”, as it turns out), is a micronutrient that’s found in every cell of a healthy human body. All eukaryotes use it, but only certain bacteria can produce it. Plants (being eukaryotes) get it from their soil bacteria, and animals get it from their gut bacteria—or from the plants and animals they eat. In that sense, it’s basically a vitamin: something that your body needs in order to function properly, but which you can’t make on your own.

Unlike the better-known vitamins, queuosine was only discovered relatively recently—sometime in the 1970s, when scientists found that certain kinds of tRNA (the “adaptor” molecules that read the blueprint of mRNA and turn it into proteins) had an unusual modification that caused them to behave differently in chemical tests. Not knowing what they were seeing, the researchers reported on this as-yet-unidentified “Q factor”. When it turned out to be a heavily modified version of guanosine—one of the four standard nucleosides that make up the rungs of the ladder-like structure that is DNA—they kept the Q, and named it queuosine by analogy.

It would be a while before they figured out what it does, and longer still until they figured out that we can’t make it ourselves, but to pack a decade or two of research down into a couple of sentences: Q is grease for the gears in the machinery that makes proteins. It sits in a part of the tRNA called the anticodon (the bit that sticks to the mRNA) where it helps that sticking happen efficiently, and—just as important—helps the tRNA “unstick” when it’s done its job and added an amino acid to the growing chain of a protein.

Our bodies have enzymes that break down RNA from the digestive tract, freeing up queuosine from food or gut bacteria. That queuosine is then broken down into a smaller molecule called queuine—the nucleobase that corresponds to the queuosine nucleoside, the same way that guanine corresponds to guanosine, adenine to adenosine, etc.

Queuine is transported via the bloodstream to cells throughout the body, where another enzyme takes it and inserts it into the tRNA in the correct spot—swapping it out for a guanine. tRNAs that are supposed to have a queuosine in them but don’t can still do their job—but they don’t do it as well as they’re supposed to.

The upshot of all this is that, in an animal that’s queuosine-deficient, a number of proteins don’t get synthesized as fast as they should, and certain cellular processes start to malfunction. It took a long time for someone to actually test this in an animal to find out what goes wrong, because inducing a deficiency of something that’s produced by common gut bacteria is a pain in the ass.

See, ordinarily, when scientists want to study something like a vitamin E deficiency, they’ll take an animal, put it on a chemically defined diet—pure fat, sugar, protein, minerals, etc.—and add in all the known vitamins except E. When the animals start to look like they’re having a rough week, maybe you add vitamin E back in to half of their diets. Whatever this seems to fix, you can conclude is dependent on vitamin E. To do the same thing with a compound produced by gut bacteria, you need to start with germ-free mice. (Think “Bubble Boy”: born by C-section, doused in antibiotics for the first few weeks of life, kept in sterile isolators, and fed a diet that’s been gamma-ray sterilized to avoid introducing incidental bacteria or fungi that might take root in their guts.)

Because these experiments are costly and resource-intense to run, it was only in 1997 that someone actually managed to run a Q-depletion experiment—but what they found was startling. On a normal diet, germ-free Q-deficient mice didn’t show any overt abnormalities—or at least, none that could easily be distinguished from the abnormalities found in all germ-free mice. (Germ-free mice are not, generally speaking, a happy bunch. They are dumber and less athletic than their ordinary counterparts, and less social to boot.) But when they took the tyrosine out of these Q-deficient animals’ diets, every one of them died within a couple of weeks.

The reason, as that 2011 paper would eventually reveal, is that Q deficiency causes a failure in the enzyme systems which prevent BH4 from oxidizing into BH2. Too much BH2, and dietary phenylalanine stays stuck as phenylalanine, rather than turning into tyrosine. The researchers behind the 2011 paper even note in their discussion section that Q deficiency “could potentially limit the production of numerous biogenic amino neurotransmitters whose production is also under the control of BH4”, but seemed content to leave the investigation of this possibility to someone else.

Of course, mice aren’t humans, so it’s not a guarantee that a Q deficiency in humans would look like what we see in mice—but in this particular case, there’s good reason to expect that the effects of a Q deficiency would actually be worse for a person than for a mouse. See, one of the major differences between our species is that humans can’t produce ascorbic acid—AKA vitamin C—on our own, whereas mice can. Ascorbate plays a major role in the biology of biopterins, helping stabilize BH4 against spontaneous oxidation and keep it active, which means that humans ought to be more susceptible than most animals to things that disrupt biopterin’s sensitive oxidation-reduction cycle.

So, to recap: It’s been proven that, in mammals, Q deficiency leads to an excess of BH2, the “dead battery” which prevents the AAAH enzymes from working as they should. We know that the AAAH enzymes are essential for the production of dopamine and serotonin, which means that a Q deficiency should lead to a deficit of these neurotransmitters and their metabolites, accompanied by elevated levels of alternative byproducts. In other words: a deficiency of this microbial metabolite would lead to a syndrome that looks an awful lot like depression.

So we have here a single-molecule, vitamin-deficiency-style model of one of the most devastating mental illnesses in the modern world.

In Vivo Veritas

But how likely is it that you’d see a Q deficiency in humans? After all, if it’s such a pain in the ass to create a deficiency in animals, you’d expect it to be equally difficult in a person—especially since there’s no such thing as a “germ-free” human. Even if someone had a deficiency, shouldn’t you be able to fix it by eating enough bacteria? We’d have noticed by now if yogurt cured depression.

But here’s the thing: not all bacteria produce queuosine. Lactobacilli and Bifidobacteria—the two genera that comprise the vast majority of fermented foods and probiotics on the market today—only have the final few steps of the Q biosynthesis pathway in their genomes, which means they can’t produce it without help from other microbes.8 What’s more, a number of bacteria (primarily pathogens, but some “pathobionts” like Ruminococcus gnavus) have a Q salvage system much like our own, meaning they act as a “sink” for Q supplied by the diet or by other gut bacteria. The same is true of fungi like Candida. If you want to guarantee a Q deficiency in an animal, germ-free is the way to do it—but it’s entirely possible that some people’s microbiomes have an imbalanced ratio of Q producers to consumers, and that over months or years this leads to a deficiency in their cells. By analogy, we can look at vitamin K—another nutrient that’s found in food and is produced by gut bacteria, but which can be depleted by antibiotics. As it turns out, 8-31% of adults tested are found to be vitamin K deficient—so a Q deficiency seems plausible.

As of right now, the hypothesis that Q deficiency is involved in psychiatric diseases is perfectly valid—meaning that it all works logically: Q is essential for keeping up a healthy BH4/BH2 ratio, that ratio is essential for serotonin and dopamine production, and those neurotransmitters are essential for mood and cognition. Whether or not it’s true remains to be seen; that can only be determined through experiment, and in that respect this hypothesis is still almost entirely untested. There’s no data on the prevalence of Q deficiency in humans, or whether deficiency is associated with any particular disease.9 We don’t know how much a healthy person needs in a day, or how that figure might change with things like antibiotic use. We don’t even know how much people get from their diet vs. their microbiome; the few datapoints we have on levels of Q in food show that a lot of foods have next to none, while some (coconut water and wheat germ) have up to a few hundred nanograms per gram.

I mentioned at the beginning of this post that I filed a patent around it—a “use patent”, covering the use of Q to treat neuropsychiatric diseases, since you can’t patent a natural molecule. That patent still belongs to my old company Holobiome, and they’ve been working on the problem—but the methods we were using for characterizing Q levels there were laborious, and required hunks of liver tissue, which means we were restricted to working with postmortem samples of dubious quality.

But I suspect there’s an easier way. Queuine, the “free” form of the molecule found in blood, is a lot like the B vitamins, in that it’s water-soluble—which means anything in excess of what your body can use in a day comes out in the urine. Testing this hypothesis might be as simple as checking urine levels in a few dozen patients and healthy controls.

If you struggle with depression and you’re anything like me, you’ve already tabbed away to google “queuosine supplement buy” or something similar. If so: welcome back, dear disappointed ones. At present, there’s no company selling anything more than research-scale quantities (i.e. $500 for a miligram, only shipping to laboratory addresses).

But fortunately, you know a guy.

See, a year or two ago, I got fed up with waiting for preliminary data and went full mad scientist. I hit up a contract synthesis company and had them make a small batch of queuine, and (after having it NMR’d and mass-spec’d by a third party), I tried some. Unfortunately, I don’t have depression or schizophrenia, and the thing about vitamins is that you only really feel something from them if you were deficient in the first place. To the guy whose gums are bleeding from scurvy, a glass of orange juice is probably better than sex—but to everyone else, it’s just a glass of juice.

But I tried it, and when I didn’t die, I gave it to some friends—and we can now say that, at the very least, queuine is safe for human consumption at doses well in excess of what it should take to fix a deficiency.

One of those friends has a sister, though, who’s been suffering from schizophrenia her entire adult life. And when she tried it, there were some remarkable changes: her family described her as more engaged and helpful, more sociable. Apparently, after a few weeks she was feeling so good that she decided to go off her other meds. This, of course, caused problems of its own, but I consider it a promising data point.

The custom synthesis cost a little over $10k, and I’ve almost run out of my personal stash, but I’ve still got contacts at the company that produced it—so if you’re interested in trying some of this stuff and have a few grand to kick in on a project like this, shoot me an email and we’ll coordinate something. Doing the urinalysis discussed above would require “heavy” standards of queuine and queuosine—versions with an extra neutron in the nitrogens—and I’ve got a quote from an academic chemist who can produce these for somewhere in the neighborhood of $20k.

Cretins

If your only context for the insult “cretin” is from its use in Monsters, Inc., you might be surprised to learn that it was originally a medical term; it referred to someone with congenital iodine deficiency. Lacking the thyroid hormones that contain iodine as part of their structure, people suffering from the condition would end up short in stature, with enormous goiters, rough skin, and severe intellectual disabilities. As recently as the 1800s, you’d see entire villages affected by it to one degree or another in landlocked places with iodine-poor soil like the Alps.

And people in these villages just accepted this as part of life: some portion of the population, they must have said to themselves, is just gonna turn out as idiot dwarfs. What can ya do about it?

This turned out not to be true, of course. With the recognition of iodine as an essential nutrient, and the introduction of iodized salt over the following century, the disease vanished within a generation—and the insult started to lose some of its sting.

So this is my hope for the future: that in a hundred years, “schizophrenic” will be like “cretin”, or “scurvy”. Once, these words meant terrible things; they were the names of demons that might curse your child, or steal your life on a long sea voyage. Those words were full of ominous power, and we robbed them of it.

If a Q deficiency is truly behind depression or schizophrenia, we might be able to do the same, simply by adding it to the salt, or the flour—the way we do with the other B vitamins. And that might be all it takes to lock these terrible demons away in the annals of history—to turn their names into anachronisms, words without any real salience, except maybe as an insult that nobody can truly get offended by, or a word that children shout, playing pirate games.

Only time and data will tell if a Q deficiency really is involved in these diseases. It’s perfectly possible that it’s not; there are probably a dozen other things just like this in the microbiome, other valid hypotheses that could explain these diseases, waiting to be discovered and tested. But if this whole saga has taught me one thing, it’s that—even if this particular idea doesn’t turn out to be the one—there’s hope in the microbiome.

It doesn’t have to be like this.

—🖖🏼💩

Subscribe now

Even so, things aren’t looking good for the endocrine system of the average modern person. Hormone-linked cancers like prostate and breast cancer are major killers. Diseases like PCOS seem to be on the rise. And for years now, scientists have been sounding the alarm about widespread declines in sperm count, along with an apparent drop in average blood testosterone levels. But in spite of all this evidence that something is wrong, we’re still largely in the dark about what that might be.

That’s why it’s surprising to me that few people seem to be aware of the studies—now more than fifty years old—showing that the gut microbiome is a key node in hormone metabolism, and that antibiotics can seriously disrupt it.

That research dates back to the early 1970s, when scientists discovered that a standard prescription course of drugs like ampicillin results in major shifts to the body’s handling of estrogens and other steroid hormones. Decades before anyone had uttered the word “microbiome”, researchers had hit on some of the first hard evidence that gut bacteria are responsible for health far beyond the confines of the gut. Now, more than half a century later, we’re just beginning to reckon with the implications of their findings.

-sterols and Steroids

In a previous piece, we discussed enterohepatic recirculation, the process where your liver secretes cholesterol and bile acids into the GI tract, to help digest and dissolve1 nutrients. Those nutrients are absorbed in the small intestine, and the cholesterol goes with them—traveling through the bloodstream back to the liver, which will spit it back out into the intestines again. Each time a molecule of cholesterol makes that journey, gut bacteria have a chance to get their hands on it, transforming it into coprostanol—a chemical that can’t be re-absorbed by the intestines. As a result, you shit it out, which helps prevent cholesterol from clogging your arteries.

But bile acids and cholesterol aren’t the only kinds of compounds that make this trip. Estrogen and testosterone are described as steroid hormones because they’re produced in the body from cholesterol, and the liver handles them in a similar way: When a steroid passes through the liver, enzymes there can conjugate it—tagging it with a sugar or a sulfur molecule to deactivate it—and then dumping it into the gut to be pooped out.

Typically, though, that’s not the end for a hormone: certain species of gut bacteria possess glucuronidase enzymes that can rip that sugary “tag” off, and this reactivates the hormone and allows it to be reabsorbed in your intestines. The net effect is that—although every molecule of steroid hormone in the body gets sent into the GI tract for disposal at one point or another—the vast majority (about 90%) of hormones that actually make it out of the body do so via the urine. In short: where the microbiome’s complex chemistry helps us get rid of cholesterol, it does the opposite for hormones—helping them stick around longer than they otherwise would.

That’s why, when scientists in the late ‘60s and early ‘70s performed experiments to see how antibiotics impact the body’s hormone systems, the first thing they noticed was that they cause a temporary but alarming decrease in urinary levels of estrogen and a variety of related hormones. Before long, though, a Finnish physician named Herman Adlercreutz would discover that this sudden drop in urinary hormone levels was accompanied by a simultaneous rise in fecal levels of the same compounds, in sugar-tagged form. From this result, he and his colleagues deduced that antibiotics don’t interfere with the production of these hormones, but with their metabolism by gut bacteria: In a healthy body, the microbiome acts as a recycling service for hormones conjugated and secreted by the liver. When antibiotics decimate our bacteria, the recycling plant goes offline, and hormones instead end up going to waste as they’re excreted in stool.

Adlercreutz’s investigations found that, in the short-term, antibiotics don’t have a substantial impact on blood levels of hormones—just on whether they come out in stool or in urine. Even so, his reports contain hints that disturbances to the microbiome could pose a risk to hormonal health: the observed increase in fecal hormone levels was smaller than the decrease in urinary ones, leading him to speculate that bacteria in the colon were destroying the remainder.2 He also found that plasma levels of androstenedione, a precursor for testosterone production in the body, dropped significantly in male subjects by the end of a five-day course of antibiotics.

Adlercreutz’s work focused mostly on estrogens, but there’s growing evidence that testosterone is subject to the same processes—which might explain why certain antibiotics are known to tank testosterone levels in lab animals, accompanied by temporary infertility. Some of Adlercreutz’s contemporaries even reported measurable drops in blood testosterone in humans taking antibiotics. This effect wasn’t seen in Adlercreutz’s studies, potentially due to differences in dose, type of antibiotic, or the initial gut health of the subjects; his cohort in the study linked above consisted of five healthy men, and none got so much as a case of diarrhea from the antibiotic.

In most cases, the changes in hormone metabolism induced by antibiotics are temporary: you go off the antibiotic, your gut bacteria grow back, and everything is working normally again a few days later. That said, the microbiome is complex and hugely variable from person to person, so what’s true for most people may not be true for everyone. There’s a prime example of this in the research on antibiotics and cholesterol: for about one person in ten, a week of antibiotics is enough to semi-permanently annihilate the capacity for coprostanol production, putting them at higher risk of heart disease. If all you’ve got is a five-person study, it’s all too easy to miss a one-in-ten effect—which means there’s a real possibility that microbiome disruptions are at the root of some people’s hormonal woes.

Estrogen(s)

Why does the body even go through this weird rigamarole of conjugating hormones and sending them to the gut, only to have its work undone when the hormones are sent back into circulation by the microbiome? The answer may lie in the fact that deconjugation isn’t the only reaction that can happen while hormones are in the GI tract.

For background: The term “estrogen” doesn’t refer to a single chemical, but rather a closely-related group of compounds that govern the reproductive cycle in women. The three main ones are estrone, estradiol, and estriol. Estradiol is the most important, being 10 times as potent as estrone when it comes to activating estrogen receptors, and 100 times as potent as estriol (ordinarily a minor hormone, but one which plays a big role in pregnancy). 

In 1980, Adlercreutz & co. reported their finding that certain Bacteroides species can transform estrone into the more powerful estradiol, while other microbes like Streptococcus can perform the reverse reaction—downgrading estradiol back into estrone. The body can carry out these same transformations on its own, but it seems likely that shuttling hormones into the gut is a way of outsourcing some of this work. This might be why some hormonal disorders like prostate cancer and endometriosis are linked to alterations in the microbiome.

But I think this kind of thing goes beyond questions of “health vs. disease”; I suspect it can explain some major aspects of people’s personalities. I recently found out that I have a bacterium in my gut which converts cortisol—the “stress” hormone that helps adrenaline do its thing—into a weird steroid derivative that prevents the synthesis of cortisol by your own cells. I’ve always joked that I have a pathological lack of anxiety,3 that I’m relaxed even when I shouldn’t be. It gets me into trouble sometimes (I’ve ended up way too high in a tree more times than I can count) but I also consider it one of my greatest strengths, and now I have to wonder if I owe it partly to this little self-limiting feedback loop in my gut that sets a cap on cortisol production.

Recent work4 has even shown that gut bacteria also have the capacity to produce neurosteroids—hormone derivatives that control the activity of neurotransmitters like GABA in the brain. One of these, allopregnanolone, was recently approved as a treatment for postpartum depression, but its short half-life means it has to be administered on an IV drip in the hospital, at a cost of something like $30,000 for a course of treatment. If gut bacteria produce this or similar compounds, they might provide a much cheaper and easier way to produce the same effect—as well as a potential explanation for why some people get PPD in the first place, if they’re lacking those bacteria.

The Takeaway

For starters, all of this is another powerful reason why everyone should be mindful of their microbes—and a potential leverage point in the fight against trends like antibiotic overuse. “Top-down” approaches like pressuring doctors to prescribe less don’t seem to work very well, in part thanks to the customer-service model of modern healthcare. If we want to curb the growing crisis of C. diff and other antibiotic-resistant superbugs, the easiest way to do it is to reduce the number of people walking into their doctor’s office demanding a Z-pack for their COVID. And the easiest way to do that is by educating people about the risks they’re taking—particularly when those pose a threat to near-universally treasured aspects of people’s identities, like sexuality and gender.

Even beyond the immediate public health implications, the microbiome is the greatest source of biological diversity among people—which is why I often say that if you’re looking for why one person is chronically ill and another is healthy, it’s the first logical place to check. But that logic doesn’t just apply to health vs. disease—it’s true of any biological variable. Knowing that hormone metabolism is under microbial control, it seems highly likely that a lot of the natural variation in things like testosterone levels is attributable to the presence or absence of certain bacteria.

Microbes are the future of health and medicine because they offer a natural, field-tested way of altering our biology—one that’s extremely simple compared to things like gene therapy. In the bacterial-utopian future, the probiotics aisle lets you pick genes a la carte: If you’re trying to age gracefully, a steroid-deconjugating probiotic might be a natural substitute for hormone replacement therapy. If you’re infertile and it turns out your estrone/estradiol ratio is out of whack, displacing some Streptococcus with the right Bacteroides might put things right again. Transitioning from male to female? Maybe a testosterone-degrading microbe could facilitate the process.

It’s a future of unprecedented agency, but it’s also a future full of hairy questions. Maybe we’ll find out that a bunch of people who want to transition male-to-female already have testosterone-degrading bacteria in their guts. “Doping” with synthetic steroids will get you banned from most competitive sports, so what happens if it turns out that 90% of Olympic medalists have a microbe that lets their body make more efficient use of their natural steroids? Should a probiotic like that be banned the way steroids are? And should people whose microbes naturally provide them with that kind of advantage compete in a separate category, or is that unfair to them?

That future is a while away, though. In the meantime, if anyone has GigaChad’s contact info, get at me. Just wanna see something.

—🖖🏼💩

P.S.: big thanks to reader Sarah Browne for catching me in a goof; the earlier version of this piece cited antibiotic use as surprisingly effective in treating endometriosis…turns out the study I was drawing on actually referred to endometritis, which is a very normal thing for antibiotics to be effective in treating.

Subscribe now

Part of what makes this a perennially effective piece of engagement-bait is that there’s a weird racial element to the discourse: washing meat before cooking is often purported to be a “black people thing”. There’s a grain of truth to this, in that white people are more likely to be “non-washers” in the US—but on the whole, ~70% of Americans report that they do it.1 Complicating the matter is the fact that people can mean a dozen different things when they say “wash”; most give their meat a quick rinse-and-rub under the tap, some use a lemon juice or vinegar solution, still others say it’s actually more like a brining process—and a few absolute madmen are apparently out there genuinely scrubbing their chicken breasts with a dollop of Dawn or other detergent.2

Now, I feel uniquely qualified to weigh in on this because in my lifetime I’ve been both a physicist and a microbiologist, and both of those domains have insight to offer here. The truth of the matter is that there are good reasons to rinse your meat before cooking—but you have to do it right.

But Cooking Kills Bacteria!

Sure. Unless they’re spore-formers.

In a previous post, we talked about the threat of Clostridium perfringens, a fast-growing bacterium that represents one of the most common causes of food poisoning. Most strains won’t give you anything worse than a case of the runs, but certain rare varieties also produce a toxin that destroys the myelin-producing cells in the brain, causing a syndrome that looks suspiciously like multiple sclerosis in animals. Suffice to say: it’s bad stuff and you don’t want it in your body—but it’s hard to avoid, because something like 80% of grocery store chicken samples test positive for C. perfringens.

The main reason that food poisoning caused by C. perfringens is so common is that the bacterium produces spores, which are like seeds for bacteria. A dried kernel of corn can survive a drought, a hard freeze, and a dozen other conditions that would kill the full-grown plant—and by the same token, bacterial spores are resistant to heat, acid, antibiotics, and most other methods of decontamination. This is why, when we’re making nutrient broth to grow bacteria in the lab, we use an autoclave: an industrial-strength pressure-cooker that can kill even the hardiest of spores. But recipes that call for pressure-cooked chicken are few and far between, so if you want to reduce your exposure to pathogenic spores, it’s hard to beat physically removing the bacteria from your food—and rinsing is a pretty effective way to do that.

It’s not perfect, of course, but it doesn’t need to be. Microbiologists often talk about pathogens in terms of the “ID50”—or “infectious dose for 50%”. You might have heard of the concept of “LD50” (or lethal dose - 50%) for toxic chemicals, e.g.: if you give 2mg of fentanyl to a group of 10 people, about 5 of them will die and 5 will survive. Similarly, the ID50 is the number of cells of a given bacterium you need to ingest for there to be a 50% chance of it making you sick. Anything that lowers your total exposure reduces the odds that you’ll get a dose that’s too much for your immune system to handle. Infectiousness depends on more than pure numbers, though: bacteria are also a lot better at surviving stomach acid and causing disease when they’re in the form of a biofilm—effectively a layer of slime made from their secretions. Gently rubbing a piece of meat while running it under the tap is usually enough to wash that slime down the drain, taking most of the bacteria with it.

Stable vs. Labile

Things like Salmonella and E. coli don’t produce spores, but spores aren’t the only nasty bacterial byproducts that can withstand cooking: there’s also heat-stable toxins to worry about.

When microbiologists talk about toxins, it’s not in the nebulous, woo-adjacent sense that can supposedly be addressed with a “detoxifying” juice cleanse. Bacterial toxins are proteins that microbes produce in order to wreak havoc on your cells, and they’re usually the main thing responsible for the symptoms of disease. Cholera toxin, for example, works by forcing your intestinal cells to spew out water and salts into the GI tract—resulting in the watery diarrhea that gives the microbe a great chance of getting into a new host.

But cholera toxin is what we call heat-labile: cooking denatures it, effectively removing its toxic properties, so it’s only a problem if you get a live colony of the organism in your body. In contrast, E. coli can produce heat-stable enterotoxins which retain their ability to give you the shits even after being exposed to common cooking temperatures. It’s not as serious as an infection with the live bacterium, where the toxin factory sets up shop in your intestines…but if your chicken has been sitting in the cooler long enough to grow a biofilm, there’s a chance that the slime contains enough heat-stable enterotoxin to cause some serious (albeit temporary) GI distress—even if you kill all the bacteria that produced it.

But Washing Splashes Bacteria!

Not if you use some goddamned common sense.

Yes, if you set a raw chicken breast in a colander at the bottom of the sink and turn the faucet on full, you will get “chickensplash”. The same is true if you blast it with the sprayer-nozzle: the amount of rebound you get depends pretty heavily on how fast the water is moving when it hits the surface of the meat. We know this for a fact, thanks to some physicists who tested splash under various conditions, and found that faucet height and water pressure were the primary factors determining how far contaminants spread.

But when you actually look at how people act in the kitchen, this starts to look like a much less serious problem than it’s made out to be. In 2019, the USDA published the results of a remarkably thorough study, where they had real people come into a test-kitchen and prepare a meal the way they would at home, using chicken thighs inoculated with a harmless “tracer” bacterium. They also had participants prepare a salad, and checked both the salad and the kitchen surfaces for contamination with the tracer bacterium, allowing a comparison of contamination rates between chicken-washers and non-washers.

The result? When real people wash chicken, the bacteria generally don’t go much further than the sink. Even within a mere six inches of the sink’s edge, the tracer bacteria only showed up 20% of the time after people washed their chicken. 12 inches away, that number drops to 8%—and the counts of bacteria were so low as to be negligible. Curiously enough, while the salad was contaminated slightly more often for washers than for non-washers (25% vs. 20%), when non-washers’ salads did have the bacteria in them, there was seventy times as much as in washers’ salads. One potential explanation for this is that, if you’re washing your chicken, you’re more likely to be conscientious about washing your hands afterward—while a non-washer might forget to give their hands a good scrub between slicing the chicken and diving into salad prep.

Best Practices

All told, there are some good reasons to rinse your meat before you prep it, and—provided you do it right—you’re not very likely to create a contaminating cloud of bacterial mess. That’s as simple as keeping the water pressure on low or medium, holding the meat a few inches from the tap, and giving it a good rub-down. If you avoid letting ready-to-eat food touch the inner surface of the sink afterward (a no-brainer), you’re practically home free.

Of course, I don’t expect this post to truly settle anything; most people arguing on the internet seem to be more interested in feeling superior than in understanding or finding truth. But at the very least, I hope it provides some reassurance for those in the civilized majority who rinse their meat before cooking it: grandma knew what she was talking about.

—🖖🏼💩

It’s uncomfortably warm to be in a jacuzzi, and I ask if it’s been long enough, if I can get out. He gives a kind of shrug which says: Your guess is as good as mine.

I get out, and take a thorough shower in the pool’s bathroom before going back over to meet him in the umbrella-shade of the table where our things are set. From his bag, he produces a box of waxing strips, which he hands to me, and a pair of gloves, which he puts on. He shows me how to apply the strips, and offers to do the first one.

Sure, I say. He yanks the strip away.

It stings, but less than I was expecting. That wasn’t bad at all, I think. I look down at my armpit. There’s still basically as much hair there as before.

Getting it all takes practically the whole box of strips—four or five tries on each side—and each subsequent yank is more painful than the last, eliciting stifled grunts of pain to accompany the distinct sound of adhesive paper being peeled from skin. The pool area is mercifully empty; the only other person there is an elderly woman swimming laps, who has apparently resolved to pretend that we’re not there. I can’t help wondering what she thinks is going on. Just one dude teaching another how to wax?1

At last, the hairs are gone. I take a deep breath and inspect my armpits, noting with interest the thousand tiny droplets of red, welling up like gems where some of the hairs had been. I am bleeding from the pores.

My companion in this adventure reaches into his bag again, and produces a bottle of ethyl alcohol and some lime juice. My stomach lurches.

Hey, alright, shots! I quip. Where’s the salt?

In my heart, I know that I will not be so lucky.

*Record scratch*, I bet you’re wondering etc. etc.

The above is not, in fact, the story of how I destroyed my armpit microbiome; it was part of the quest to fix it. But we’ll get back to that. First, let’s rewind another three summers.

It’s 2014, and I’ve just graduated college. I’m living, for the summer, in the dorms of George Washington University in DC—courtesy of an internship with the Society of Physics Students.

It’s early in the day, but already clear that it’s going to be a stifling one. Outside the bars of my bathroom window, a busker with a saxophone is riffing on the Star Wars theme. I hop out of the shower, hurrying through my morning routine so I can catch the metro out to College Park, where the American Physical Society has its headquarters. I go to put on deodorant, only to wince as the plastic scrapes my pits; the stick I’m using has reached the end of its life.

I’d known this was coming—you never reach the end of your deodorant without being aware at least a few days in advance—and I curse myself a little for failing to get ahead of the problem. But then an idea strikes me. I remember reading somewhere that sweat itself isn’t actually responsible for the smell of body odor—rather, it’s bacteria living in the armpit, which eat the fatty acids in your sweat and excrete odorous byproducts.

Like any good intern fresh out of college, I had my priorities straight: no deodorant, but plenty of alcohol in the house. I even had some homebrew limoncello in the works, which at that point was just a bottle of vodka with a few lemons’ worth of zest steeping in it.

Hey, alcohol kills germs, I say to myself. If it’s the bacteria that smell, maybe this’ll do. I splash some of the lemon-scented vodka onto my hands and rub it into my armpits. I congratulate myself on my ingenuity and—lemony fresh—set out for work.

And it works! Despite the muggy summer heat, I go the whole day without catching an offensive whiff of myself, and don’t even think about the deodorant problem again until the next morning, when I’m faced with the same dilemma as the day before. I do the lemon vodka pat-down again, telling myself that I should probably go and get some actual deodorant sooner or later, but still pleased that I’ve found an interim solution that works so well.

After three or four days of this, I’m starting to seriously cut into the yield on this batch of limoncello. I skip the alcohol rinse and head for work, resolving that tonight is the night I’ll get my shit together and run to the store for some deodorant.

Halfway through the day, I realize: I have made a terrible mistake.

Body odor is one of those things that everyone seems to experience differently, but hopefully you understand what I mean when I say that most people’s sweat shares a characteristic undertone of funk. It’s a little spicy—somewhere near the smell of cumin, in olfactory space—but it’s not exactly unpleasant. On the right person and under the right circumstances, it can even be sexy. Being trapped on a train car with too much of it can start to get claustrophobic, but it’s fundamentally a sweaty human smell.

On this first day that I went without deodorant, I discovered that this smell—any semblance of a normal human body odor—was gone from my pits. In its place was what I can only describe as metal and onions with a side of mildew, although that doesn’t quite capture it. It was sharp, rank, and dank in all the wrong ways. What the Grinch would smell like, if you could smell him.

I was already aware of the concept of a microbiome by this point, thanks to an evolutionary biology class I had taken in my final year of undergrad. I knew that these ecosystems were often fragile, and could be tipped out of balance permanently by a single perturbation. I even had a strong hunch—bordering on a conviction—that such disturbances in the gut were behind a number of psychiatric diseases.

So it didn’t take long for me to realize what I had done: I had burned down the jungle. Whatever innocuous microbes had been living all my life in the verdant rainforests of my armpits, fermenting my sweat into those normal-funky-human byproducts—they were no more. Instead, all I had was whatever hardy bastards had survived the fire, or else grown back quickly enough to take over.

That night, after a train ride spent deliberately sitting as far from anyone else as possible, I bought a stick of deodorant—one of the crunchy, “natural”-branded ones, with some kind of vegetable wax as its main ingredient. I wanted something to keep the stink down, but was worried about doing even more damage to the ecosystem than I had already done. It did a fine job of keeping the problem in check. As long as I didn’t smell at all, I didn’t smell bad. I would check, periodically, to see if the situation had righted itself: Right out of the shower, when the previous day’s deodorant had been washed off, I would lift and sniff.

Bad. Bad, bad, bad, even fresh out of the shower.

A Visit with Dr. Armpit

Fast-forward to 2017, to the sunny poolside where we started this story. I’m in San Diego on a work trip, and—after years of experimenting with various remedies both natural and artificial—I’ve managed to find a researcher who studies the axillary2 microbiome for a living, and happens to live in San Diego. He’s sympathetic to my situation, having found his way to the field through a story similar to my own: apparently, he had picked up a bad Corynebacterium from sharing a bed with a rather smelly girl. Still, he’s not keen to experiment on a random stranger from the internet. It takes pleading, cajoling, and—eventually—the promise of a few hundred dollars in cash to convince him to devise a protocol.

When we meet up, he does a quick sniff test and confirms that I’m likely dealing with the same type of organism: Corynebacteria are a part of most people’s axillary microbiome, but they’re usually a minor component, kept in check by native Staphylococci and other less malodorous microbes. It’s a hard problem to rectify, he explains, because these organisms dwell in some of the deepest layers of the skin—down near the base of the hair follicles in the tiny glands where sweat, oils, and a waxy secretion called sebum are produced.3

This is why the first step is a soak in the hot tub: to liquefy the waxy stuff at the base of my hair follicles and make it easier to clear them out. After the decidedly unpleasant waxing process, all that’s left is to take our best shot at sanitizing the turf, and then reinoculate with good bacteria.

Easier said than done. If the waxing had stung, the next part seared. As I swab the area with alcohol on a sterile gauze pad, I become exquisitely aware of just how sensitive a body part the armpit really is. My skin flushes hot, as alcohol mixes with the blood still seeping from thousands of freshly opened follicles—flowing into the tiny channels like liquid fire, then like ice as the fluid starts to evaporate and cool the skin. My nerve endings scream out to my brain in alarm and confusion. I swear through gritted teeth. I take a breather before doing the other side.

The lime juice is next. It seems like a cruel joke, and if I hadn’t initiated this whole thing I would wonder if I’m being pranked—but the man who devised this new and interesting form of torture explains to me that low pH plays a crucial role in biasing the environment against Corynebacterium and in favor of the “good guys”. By this point, either the nerves have given up on their distress signals, or the series of assaults has triggered such a rush of endorphins that I barely notice as the juice goes on.

The worst of it is over.

A little dazed but rapidly recovering, I watch with interest as my companion reaches into his bag and pulls out a stack of petri dishes wrapped in a waxy film.

Hesitantly, I ask: Are you sure we can’t just…I make a finger-twiddling gesture toward his armpits, then toward mine, accompanied by a little boodleoop noise. He gets a grave look on his face. Absolutely not.

In response to my disappointed expression, he elaborates. You’ve heard of MRSA?

I had. It nearly killed my mom, after a surgery a few years prior.

Better to be stinky than dead, right?

I had expected this. In our email exchange, he had explained to me that, with a complex community, there’s no telling what will take hold and what won’t. A transplant would run the risk of introducing an opportunistic pathogen like MRSA—while failing to transfer some other species that would prevent it from becoming problematic. Instead, the plan is to introduce one well-studied commensal called Staphylococcus epidermidis; a dominant species in the axillary microbiome of people who have an inoffensive body odor. There’s no way of knowing if it’ll work, but it’s a best guess—and if it goes wrong, we know which antibiotics can kill it.

I accept it, nodding as I lean back against the rubbery straps of the poolside chair. He carefully unwraps the plates, and through the clear plastic lid of the petri dish, I can see a beige substance stretching uniformly over the surface of the gelatinous material inside. He punches a sterile cotton swab out of its package, and runs it through the beige gunk with a rolling motion, picking up as much of the stuff as possible.

We inoculate my armpits.

The Way Back

The following week is a weird one. I work our booth at Comic Con, handing out physics-themed educational comics and little LED circuit kits to anybody who will take one. Under my shirt, the makeshift bandages chafe whenever I move my arms: we’ve rigged up an elaborate system of gauze pads and medical tape to keep as much of the inoculum on my skin as possible. I take them off to shower—no soap on the pits—and then put them back on with fresh tape.

By the time we’re packing up the booth to head home from San Diego, there’s a downright bizarre smell coming from the gauze. Almost cheeselike. The smell from my armpits themselves has maybe changed slightly—there’s definitely a note of the gauze smell in it now—but it’s by no means “normal”.

On the flight back east, I console myself with the knowledge that, on the ranked list of “craziest-smelling people at Comic Con”, I wasn’t even in the top quartile. At the end of the week, I trash the bandages, and the cheesy smell goes away overnight.

Slowly, my armpit hair starts to regrow (although I swear it’s less thick than it used to be, to this day) and I get on with my life, accepting that there are worse fates than a life of mandatory deodorant. Casual attempts to convince friends to let me steal their armpit bacteria go about as well as you’d expect. A few years later, recounting the story to an Uber driver earns me a laugh…and a worthwhile tip.

Where I come from, he says in a thick Ghanian accent, we just use limes.

Subscribe now

I try it that night, and am surprised to discover how well it works: the smell disappears almost instantly. I recall “Dr. Armpit’s” comment about the pH preferences of different bacteria, but suspect this has more to do with chemistry than biology because of how quickly it happens—maybe changing the pH reduces the volatility of some of the smellier compounds. But over the next year or two, as I sub in lime juice for deodorant whenever life allows it, the smell seems to gradually improve even without the juice—and I wonder if I’m reshaping the ecosystem of my armpit microbiome.

But earlier this year, I came across a paper from 1953—one of the original treatises on bacteria and B.O.: Axillary odor; experimental study of the role of bacteria, apocrine sweat, and deodorants by Shelley, Hurley, and Nichols. It’s a scientific tour de force—the kind of exhaustive investigation that, these days, you rarely see outside a big-name journal like Nature or Science. They trained a team to perform blinded olfactory ratings, tested different deodorants, injected people with epinephrine to activate their apocrine glands, collected sterile sweat, inoculated that sweat with various bacteria to explore their impact on the odor profile, tested topical antimicrobials for their effects on B.O…they really left no stone unturned.

One key takeaway from their findings: there are some people with armpit bacteria that make them smell better than the rest of us.

But the big finding—the one that made me smack my forehead? There’s a simple way to take the smell out of even the stinkiest sweat.

It’s aluminum. Normie deodorant.

It looks like this effect is also as much chemistry as it is biology, in that it’s an instantaneous change rather than one that reshapes the microbiome to produce less offensive odors to begin with. Still, it took me by surprise to realize I had succumbed to the naturalistic fallacy—assuming that “artificial” aluminum deodorants would necessarily make the problem worse.

I switched from my hippie deodorant to Degree, and—wouldn’t you know it—smelled more or less normal again within the hour. Sorry, Tom of Maine.

I still have hope that someday we’ll be able to figure out a natural solution: to find those happy few people blessed by nature with “pleasant” B.O., and convince them to share their gift with the world. It seems like the kind of product that would kill among a certain subset of the Silicon Valley crowd. It won’t necessarily be easy, though…the one genus reported by Shelley, et al—Klebsiella—is an opportunistic pathogen, frequently antibiotic-resistant and with a tendency to cause sepsis if it gets into your bloodstream. Getting clearance to sell it as a cosmetic seems like an uphill battle.

There are other layers of intrigue here, as well. Take the famous “sweaty t-shirt test”—an experiment which found that people can smell one another’s genes. The study involved giving a number of men new t-shirts, and having them get ‘em nice and gross by wearing them for a few days. Then, a panel of women were asked to huff the shirts and rate the smells on their attractiveness.

Sequencing the genomes of the participants revealed a consistent pattern in the results: a woman was more likely to rate a shirt’s odor as attractive when it came from a man whose genes were dissimilar from her own—specifically in a region of the genome called the major histocompatibility complex, or MHC. This is an immune locus, the bit of the genome which helps your body differentiate your own cells from those of other organisms. Curiously enough, it’s also where you find most of the risk genes for diseases like schizophrenia.

But we know for a fact: it’s not the sweat itself that smells. It’s the fermentation products of that sweat…so the results of the sweaty t-shirt test must be mediated by the armpit bacteria of the participants. This makes sense: if the MHC encodes the list of which cells get a pass from the immune system, it likely plays a role in determining which bacteria are allowed to stick around. One implication of this is that there may be no one-size-fits-all bacterial fix for B.O.; it’s possible that, if I tried inoculating my armpits with the Klebsiella described in that 1953 paper, my immune system would destroy it. Hell, maybe that Klebsiella only smelled pleasant to the researchers because they had MHCs very dissimilar from the people the bacteria came from.

As for what it means for schizophrenia…I have a theory. We’ll get to that next time.

-🖖🏼💩