What Should We Make Of Sasha Chapin's Claim That Taking LSD Restored His Sense Of Smell After COVID?

I.

Substack blogger Sasha Chapin writes that COVID-19 Took My Sense Of Smell, LSD Brought It Back. He got coronavirus, and like many people lost his sense of smell (medical term: dysosmia or anosmia). Ten days after recovery, he still couldn’t smell anything. He looked on Twitter and found some anecdotal reports that psychedelics had helped with this, so he took LSD and tried to smell some stuff while tripping. He says it “totally worked. Fully and near-instantaneously. Like a light switch turning on.” The details:

My idea was that I’d do some scent training while on LSD, to—hand-wavey lay neuroscience incoming—stimulate whatever olfactory neurogenesis might occur. Before tripping, I laid out my fragrance collection, along with a few ingredients from the pantry. All-in-all, there were about fifty things to smell, and, as the LSD started kicking in, I started making my way through the selection.

At that moment, my sense of smell was still somewhat there but mostly not. However, something odd was happening; I could detect some of the fragrances’ nuances that I couldn’t pick up earlier that day, and what I detected shifted from moment to moment.  It was like I was listening to a piece of music with random instruments dropping in and out of the mix. This was still a kind of anosmia, but a different kind, and it almost felt as if my olfaction was re-negotiating reality in real time.

And then another weird thing happened. For a couple of hours, I got acute short-term parosmia (distorted smell.) My nose felt dry, and a weird puke-y smell filled my mind. According to some research I’d done, in anosmic patients parosmia sometimes precedes recovery, so, though this was quite unpleasant, I felt hopeful that this was some part of the regeneration process. I cleaned the house, my wife took me shopping, we went to Home Depot, and then had dinner.

We got home soon after, about seven hours after my trip began, and I returned to my fragrance collection. Cue triumphant music: all of them were now smellable, in high-definition. My anosmia was gone. Moreover, some were more pleasant than before; iris was more palatable to me than it ever had been. This was a moment I won’t soon forget. Some fragrances—especially Dzing!—gave me full-body chills.

The next day, my sense of smell was still there, but it fluctuated; it was partial in the morning, then full in the evening. Since then, it’s been back basically 100%. (And the improved understanding of iris has persisted.)

The number one explanation for incredible Internet medical stories is always “placebo effect”. Number two is “coincidence”, number three is “they made it up”. All of these top the list for Sasha’s experience too.

Still, enough people have said something like this that I think it’s worth trying to figure out if there’s any plausible mechanism.

II.

Anosmia sucks worse than you would expect. For one thing, smell is linked to taste, so most things taste bad or weird or neutral. For another, it’s correlated with much higher risk of depression, and some preliminary work suggests this could be causal (possible mechanism: the brain is getting fewer forms of stimulation?) Some studies find that exposing rats to very strong scents makes them less depressed; it would be funny if this was how aromatherapy worked in humans. So COVID induced anosmia is actually a serious problem.

According to annoying people who refuse to provide useful information, between 3% and 98% of people who get coronavirus lose some sense of smell. A meta-analysis that pools all these studies gives a best estimate of around 40%. Lots of respiratory viruses cause some smell loss when they infect your nasal passages, but coronavirus is worse than usual. Milder cases cause more olfactory problems than more severe cases, suggesting that the immune response is at least as involved as the virus itself. The coronavirus cannot infect neurons directly, but might infect other cells in the nose, including cells which support neurons and help regenerate the olfactory epithelium.

About half of COVID patients recover their smell in a few weeks, but some cases linger for up to a year. By the end of a year 95%+ have recovered; given that between 3% - 12% of people have random smell disturbances at any given time anyway, I interpret this latter figure less as “some people never recover” and more as “we reach the point where it’s impossible to distinguish from background problems”.

Sasha says he was only ten days in when he took LSD, so this is well inside the window where we would expect him to eventually recover anyway. But it still doesn’t make sense that he recovered within the space of a few hours, or that he felt his smell was stronger than before.

III.

Sasha writes that he was “do[ing] scent training while on LSD, to—hand-wavey lay neuroscience incoming—stimulate whatever olfactory neurogenesis might occur.” He’s pretty up front that he’s using “neurogenesis” loosely, but I want to stress that probably neurogenesis isn’t a good explanation here.

First, neurogenesis probably doesn’t happen in adult humans. If it does, it’s rare, limited to a few areas, and too specific for something like LSD to affect it much. Second, even if it did happen, I would expect it to take more than the few hours that Sasha said it took him to feel an effect. Third, coronavirus doesn’t seem to directly damage nerve cells anyway.

There could potentially be synaptogenesis - existing nerve cells making new synapses - but I don’t think this is involved either. Olfactory cells have long and complicated synapses that have to pass through various bony plates; although I’m not an expert in exactly how this happens, it would surprise me if they could just sprout a new one and get it to the right place within a few hours.

If Sasha’s experiment really worked as quickly as he says, that suggests his brain was previously getting olfactory signals, but ignoring/misinterpreting them.

Steven Byrnes, who I trust a lot on issues like this, writes:

Let me see if I understand this well enough to flesh it out. The olfactory cortex gets signals from the nose and does some computational function to them to extract signal. If the nose is damaged, the previous function doesn’t work, and it needs to use a new function that (for example) maximizes the weight it ascribes to healthy cells and ignores damaged ones. But the olfactory cortex isn’t in the business of reweighting its computations on the fly - in healthy adults, it’s been doing this for decades, it’s already got a really good idea what the best function is, and if it changed its parameters for a specific situation, it would probably be overfitting to that one thing at the cost of general ability. LSD “increases the learning rate” in the sense of telling the brain that actually it should be more plastic and update its machinery more dramatically. This is usually bad (because it will make you overfit). But it can be good if your previous function is obsolete for some reason (eg your incoming sensory signals are in a different configuration than they were last time you optimized the function). Sasha took a drug that raised his learning rate, then concentrated on trying to smell things as hard as he could while his olfactory machinery was unusually plastic. That caused him to re-train his olfactory machinery for the new situation.

(does LSD actually increase learning rate? I think this is Steven’s interpretation of Relaxed Beliefs Under Uncertainty - we’ll return later to whether it’s accurate)

Steven says the learning rate is set to ~0 in adults, but I don’t know if this is true. After all, everyone (or at least 95% of people) recover their smell after COVID eventually. One way of thinking about this is that the nose finally heals, all the sensors are back in their regular position, and the olfactory cortex’s original computational function becomes relevant again. But another possibility is that the learning rate is somewhere very slightly above zero, and in a few weeks or months you manage to update your function the same amount that Sasha updated it in a few hours with his accelerated learning rate on LSD. Certainly this is how you would interpret eg the neuroplasticity that lets stroke victims eventually use their legs again even if the brain areas originally handling leg movement were irrecoverably damaged (though the brain might deliberately increase its own learning rate above normal adult levels after a stroke too).

You can find another interesting example of this process in Fanciullacci et al 1977, who find that a single dose of LSD can cause “striking” improvement in phantom limb pain. This makes sense, right? Your sensory cortex learned some function mapping inputs to perceptions back when you had your limb. Then when they amputated your limb, it never updated, and you feel the incorrectly-mapped sensations as phantom limb pain. Increase the learning rate and your sensory cortex will retrain itself on existing data and hopefully realize that the part of the model corresponding to the missing limb isn’t helpful anymore.

Sasha asks:

Questions that occur to me include: is this COVID-specific? What if LSD can regenerate other sensory deficits? What if we can make ourselves into super-tasters through psychedelic scent training? Inquiring minds wish to know.

If Steven’s model is right, we can at least answer the last (and let’s face it, coolest) of these questions: probably we can’t costlessly make ourselves into super-tasters here. The olfactory cortex is already doing its best trying to learn a function extracting signal from its sensory inputs. If you increase the learning rate, you can update the function to current conditions faster, but you can’t make it better than a normal human would come up with given an infinite amount of time.

(one exception: in theory, it might be possible to use overfitting to your advantage. Suppose you were a wine taster who has to identify many different kinds of wine by smell. If you took psychedelics and then spent your trip smelling wine and nothing else, you might convince the olfactory cortex to slightly rewire itself for the specific problem of smelling wine in particular. This would make you worse at everything else, but maybe a little better at the one task you’re optimizing for. I do not recommend this.)

IV.

I started with Steven’s explanation because I think it’s simple, powerful, and he knows more than I do and is more likely to be right. But I also want to go over the longer, more complicated, and more theoretical way I was thinking about this before I read his comment. Then we can try to figure out whether I’m looking at the same thing from a different direction, or whether these are competing models.

COVID-induced anosmia makes a lot of sense for the few days or weeks when you’re actually fighting the virus. Most people have it during that period, then get better. A few people have it during that period, then mysteriously don’t get better for many months, even though the virus is long gone. This makes me wonder if persistent COVID-induced anosmia fits into a class of problems which don’t 't have an official name, but which I think of as "state fixation disorders". This will be easier if I just give examples:

- Mal de debarquement: you know how if you go on a boat, you might get seasick and nauseous for a few hours/days until you get your "sea legs", and then you're fine? And you know how, once you get off the boat, you might feel sick and nauseous for a few hours/days until you get your "land legs" again, and then you're fine? In some extremely small percent of cases, you never get your land legs and you feel sick and nauseous forever. You can take one boat trip, get off it, something goes wrong, and you feel nauseous for the rest of your life. This is the purest example of what I mean by "state fixation disorder" - you have a state (like seasickness) that you ought to adjust away from quickly, and somehow it becomes fixed and lasts forever for unclear reasons.

- Hallucinogen persisting perceptual disorder: this is when you take a hallucinogen like LSD and then instead of the trip ending when the drug wears off, certain aspects (usually visual hallucinations) continue forever. The full-blown version is rare (though not vanishingly so; I'm treating a patient with this right now), but very mild persisting hallucinations seem to be pretty common. Like the seasickness example, a state which should be temporary (ie last only as long as the drug is in your system) becomes unnaturally fixed.

- Post-acute withdrawal syndrome: if you take a drug like Xanax a lot, you become addicted. Then if you stop it suddenly, you go into withdrawal. If you stop a high enough dose of Xanax suddenly enough, this withdrawal can be really awful, including things like seizures, hallucinations, and maybe death. But in theory, it ought to last a week or two while your body adjusts to not having any Xanax in it anymore. In most people, this is exactly how long it lasts. In a very small number of people, some of the symptoms seem to last forever. I saw a patient with this problem and it was heart-breaking. My first instinct was to put her back on Xanax, then try withdrawing again, but this didn't really work, and there are many case reports of it not working. Sometimes the patient isn't even better while they're on their full previous dose of Xanax! It seems like some part of the withdrawal process has become "the new normal" and the body is determined to stick with it in a way totally unrelated to whether they're withdrawing or on the drug or whatever. Sometimes this also happens with SSRIs, although SSRI withdrawal can be complicated enough that I am less sure there's no simple biological explanation.

- Chronic pain: what, you thought this was all going to be cool exotic syndromes you've never heard of? Chronic pain is exactly like this. You have some acute injury - maybe you fall and break your leg. It predictably causes leg pain. After a few months, your leg is completely healed. But maybe the pain continues anyway for the rest of your life. It's tempting to think that there's some microscopic injury that persists even after everything else looks normal, and for all I know maybe some chronic pain is like this. But studies find that the risk factors for chronic pain are less about the kind of injury and more about psychological factors like depression and trauma. My guess is this is another state fixation disorder; a state of pain which is temporarily appropriate for a limited injury somehow gets hard-coded in and becomes the new set point.

Chronic pain is by far the best-studied of these, and the going theory is "central sensitization" aka "nociplastic pain". Wikipedia describes it as "a dysfunction of the central nervous system whose processing of pain signals may have become distorted or sensitised".

I think this makes most sense in a Bayesian context: all of your perceptions are a combination of actual sensory signals, plus processing by a complicated analytical machinery which interprets it through a collection of priors about what it expects to be true. Suppose you break your leg and are in a lot of pain. Probably for weeks and months you get a bunch of leg pain signals. You learn to expect leg pain signals. If your nervous system is inappropriately plastic, that expectation might become one of the very high-level priors you use to interpret sensory information. So when the nerves in your leg send some complicated stream of electrical impulses plus random noise, the interpretation centers in your brain "know" that your leg "should" feel pain and are more likely to include pain signals in the "final report" they send higher brain areas.

One of my favorite teaching aids for explaining this process. Concentrate on either the phrase “brainstorm” or “green needle”, and you will hear the device saying that phrase (and not the other). You’re essentially “suggesting” an interpretation to your auditory processing centers, and they use your suggestion in deciding what perception to send to higher brain areas. A prior on chronic pain is essentially a similar “suggestion” for central pain gating areas.

My guess is that all the state fixation disorders are like this. You're in some unusual state. Your brain develops a prior that you're in the state, which is correct and probably helps you adjust while you're in it. But the prior is inappropriately strong; maybe it's even the dreaded 100% probability prior, which is impossible to ever update away from no matter how much evidence you get to the contrary. So when the state ends, your prior doesn't, your nervous system still expects to be in the state, and it interprets all of your perceptions accordingly.

LSD weakens cognitive priors. At least this is my interpretation of the Relaxed Beliefs Under Uncertainty model, probably the leading theory for how psychedelics work. If you screw up and get a 100% probability prior that you are on a rocking boat, no amount of being on land will ever be able to update it (since 100% probability priors by definition can never be updated). But if LSD weakens all priors including that one, then maybe it’s only a 90% probability prior that you’re on a boat, and you can look around and notice all the evidence that you’re not on a boat and update.

This suggests that psychedelics should be able to help state fixation disorders. As far as I know this hasn’t been formally investigated. Some people try taking psychedelics for chronic pain, and it seems to work okay, but there are a lot of people who really love psychedelics and like telling stories about miraculous results, so much so that it’s hard to tell if there really are any miraculous results here (though see the story of the woman who messed up, accidentally took the highest LSD dose in recorded history, and got away with nothing worse than it curing her chronic foot pain). Obviously no one would give LSD for HPPD, you would have to be crazy to try this. Mal de debarquement is too rare to really investigate. So I don’t really have a good sense of this.

We can speculate further if we’re okay stretching the idea of a state fixation disorder. In some sense, doesn’t PTSD belong in this basket too? You get in a state of being traumatized which is appropriate for your traumatic experience, but then it doesn’t go away afterwards, right? And we know that psychedelic therapy can be extremely effective for PTSD. Also, is a state fixation disorder the same thing as a trapped prior, except at the low level where things “feel” neurological instead of the high level where they “feel” cognitive? Not sure.

How does this relate to Steven’s idea of increased learning rate? In the state fixation disorders, it seems like at one point the learning rate was too high - eg in mal de debarquement, you had high enough learning rate on the boat to retrain your balance-related centers to predict the immediate on-the-boat stimuli, instead of keeping the generally successful function they’d used for most of your (land-based) life. But this unusually plastic condition is temporary; once you’re back on land, your learning rate is back to near-zero, and you can’t retrain to predict the land-based stimuli.

Likewise, in the people who suffer from COVID-induced anosmia for many months, you can think of them as “learning” to accept their deranged or non-functional olfaction as normal. Then after they recover, they can’t “re-learn” the correct function mapping sensory input to high-signal smell perception, so they just stay like that for a while. LSD loosens their prior on the aberrant condition and allows the system to re-learn how to process normal inputs.

I’m embarrassed to admit I don’t know enough neuroscience/machine learning to know whether “strong fixed priors” and “low learning rates” are the same thing. It seems like they might not be - I can have a high prior that any individual perception is (eg) of rocking back and forth on a boat, but still be extremely open to updating once I find that it isn’t. My guess is that these aren’t definitionally the same, but that one tends to contribute to the other through the mechanism I discuss in Trapped Priors As A Basic Problem Of Rationality.