Great writeup. I wonder if anyone has ever tried injecting anti-amyloid antibodies directly into cerebrospinal fluid (intrathecal delivery) to bypass the blood brain barrier.
Also, regarding γ-secretase modulators, a lot of work went into these in the early 2000s and they ended up making things even worse due to side effects. γ-secretase is involved in processing a lot of proteins, notably including Notch, and it's quite difficult to block APP cleavage selectively.
I'm not aware of any intrathecal amyloid antibody delivery in humans.
Regarding γ-secretase, yeah, so those were γ-secretase *inhibitors* and they did have a lot of off-target effects, because of the enzyme's large number of substrates. However, despite a wide variety of mutations, the Alzheimer-causing ones strongly negatively correlate age-at-onset with the number of large amyloid species they lead to. [1] So it's being looked at again, but via more precise targeting of just the amyloidogenic pathways, hence the term "modulator".
[1] Fernández, Sara Gutiérrez, et al. "Spectrum of γ-Secretase dysfunction as a unifying predictor of ADAD age at onset across PSEN1, PSEN2 and APP causal genes." Molecular Neurodegeneration 20.1 (2025): 48.
>I wonder if anyone has ever tried injecting anti-amyloid antibodies directly into cerebrospinal fluid
That was my first thought also. If I'm faced with the probability of dying a horrible death due to dementia, you can go ahead and inject whatever the heck you want into my spine using whatever experimental methods you want. It can't be worse than the alternative, right?
> injecting anti-amyloid antibodies directly into cerebrospinal fluid
I wonder if a better approach might not be nasal administration of amylophagous protists. I believe this has successfully been demonstrated as an effective route, with neti pots and "brain-eating amoebae."
These phase 1b tau trials are interesting. They were not designed to measure clinical efficacy but still it would have been nice to see some hint of it. Apparently we were not so lucky:
"Performance on functional, cognitive, psychiatric and neurologic clinical outcomes slightly declined as expected for participants with mild AD over the duration of the treatment and post-treatment periods." [116]
Had a chance to spend some time with that paper today, but only a few hours, and I'm otherwise not super familiar with lithium/Alzheimer research (sadly, even with all the time I've spent, the field is so sprawling that there are plenty of topics I've only scratched the surface of) so take this with a (micro)dose of (lithium orotate) salt, but here's my understanding of their claimed results and my reaction:
- In both amyloid and amyloid+tau mouse models, many lithium salts (but not lithium orotate) co-accumulate with amyloid plaques, in a manner which appears competitive with other cortical locations, so that the presence of amyloid plaques seems to result in a reduction of lithium in cortex (but not a reduction in non-cortical lithium) relative to normal mice. [One caution so far: there's a LOT of crap found in amyloid plaques. So finding lithium there may or may not by itself be of significance.]
- Human Alzheimer patients show lithium concentration in amyloid plaques, and reduction in cortical lithium (but not elsewhere) relative to non-Alzheimer humans, similar to the mouse models. (It's not totally clear to me if the "Alzheimer" patients all actually have Alzheimer pathology as opposed merely to clinical symptoms. I'd like this clarified.)
- In mice, a lithium diet deficiency accelerates the emergence of Alzheimer-like pathology and cognitive deficiencies in mice already predisposed to them, but even a small amount of lithium orotate (but not lithium carbonate, which binds better to amyloid) is protective.
So there are two potential causal pathways they're identifying:
(1) Amyloid pathology (known to be in part due to amyloid overproduction due to the mouse models being used) -> less lithium available to cortex.
The purported mechanism here seems clear: the lithium is preferentially binding to the amyloid plaques rather than residing in other cortical tissue.
(2) Less lithium available to cortex -> more amyloid and other Alzheimer pathology.
In *this* case, I'm confused about what the purported mechanism would be. I can understand how more lithium with amyloid affinity could cause more amyloid pathology (by promoting further cross-aggregation), but what they're showing here is subtly different: it's that a *deficiency* of lithium that *doesn't* have amyloid affinity is causing more Alzheimer-like pathology. Perhaps it's an interaction with a third, unspecified molecule, pulling it away from amyloid and preventing the promotion of aggregation? Would be good to understand this better.
Theoretically, it seems like there could also be a feedback loop between (1) and (2), since they seem to run in opposite directions, but I'm not sure.
A few questions I'd like answered:
- What is the timeline of cortical and amyloid-associated lithium in humans with preclinical Alzheimer's disease (i.e. with amyloid pathology but not yet at the symptomatic stage)? They didn't look at this, but knowing the answer might help to rule out hypotheses about causality.
- What is the purported mechanism by which a cortical lithium deficiency induces Alzheimer-like pathology? This sort of question can be pretty hard to answer because it's common to identify candidate molecular pathways in vitro or in animal models that end up not being the important ones in the human disease.
- Do many human beings actually suffer from a (non-amyloid-binding) lithium deficiency in our diet? If so, then that increases the relevance of these findings, at least for some of us. If not, then they may have just uncovered a theoretical pathway to amyloid pathology and Alzheimer's disease which isn't relevant to most human beings.
I don't know if it will go anywhere, but it seems interesting and worth exploring more. There were some previous Alzheimer lithium clinical trials which mostly don't seem to have panned out, but they used different lithium salts which bind better to amyloid than lithium orotate, so it's possible that this would achieve different results. Apparently they are investigating the appropriate dose range for human beings, which currently isn't clear, before pursuing clinical trials.
In the section of the paper titled Lithium and Microglial Function, they describe transcriptional changes to lithium-deficient microglia, which are immune cells that usually help clear amyloid beta from the brain - the last line of that section summarizes their findings as "Li deficiency leads to a reactive pro-inflammatory state and impaired Aβ clearance".
So I think at least part of the causal pathway here is something like: 1) amyloid starts to build up and form plaques, which start to entangle lithium and reduce cortical lithium levels, which 2) impairs microglial function, reducing their ability to clear amyloid (and increases inflammation), leading to more plaques, and so on in a vicious feedback loop.
Are there any of these drugs you would take yourself in (say) your 50s in an attempt to slow down the progression to Alzheimers? It seems like the endgame of this research is likely to be that approximately everyone 50+ takes an anti-amyloid drug along with their statin, in order to delay the onset of mental decline from tau long enough to let them die of something else. And if that prevented most cases of Alzheimers and didn't have terrible side-effects, that would be a really good outcome.
If I were amyloid-positive (as determined by plasma, cerebrospinal fluid, or PET scan) and pre-symptomatic, I'd try to enroll in the trontinemab prevention trial, which looks like the best bet right now. https://www.roche.com/media/releases/med-cor-2025-07-28
Trontinemab looks to have higher efficacy and much better safety than the on-market state-of-the-art.
I was going to ask this question too. I have one of the APOE mutations and I have a PET scan from a couple of years ago as a result of a cancer diagnosis experience.
Would it make sense I wonder to ask someone to look at my PET scan through this lens and then if there is evidence, would that qualify me for a prevention trial?
Your PET scan wouldn't have been with an amyloid tracer, so it won't be helpful here. If your only genetic risk factor is ApoE4, you're very unlikely to be in the preclinical phase until at least your late 40's or early 50's. If you are of that age, you could probably get a blood, CSF, or plasma test for amyloid pathology and then if that's positive, attempt enrollment in one of the prevention trials I mentioned (which will also confirm the pathology before enrollment).
Thank you for clarifying that. I'm 60 actually. My brother has Lewy Body dementia, but I gather that variant is not genetic in terms of raising my risk.
Sometimes the term "genetic" can be used confusingly to just mean autosomal dominant (one copy guarantees the disease), even if there are risk alleles. ApoE4 is also a risk factor for LBD, but I know much less about it.
I work in the field and just want to second David's advice here. If you're 60 and you have an ApoE4 gene there is a good chance you have pre-clinical AD, and getting an understanding of your amyloid levels now (and really every year from now to monitor the change if you're not ready to take potentially risky treatments) will be beneficial.
If I could afford it I would have a PET scan. If not, probably CSF although a spinal tap is hardly much fun. The plasma tests may not be accurate enough for monitoring yet.
Omroth, I really appreciate your taking the time to add your thoughts from your expertise.
Because of this post and your and David's comments, I've reached out for a referral to a neurologist to talk about all this, so thank you very much for the nudge. A yearly PET scan given my radiation load (BC cancer treatment) and risk of cancer recurrence sounds risky, but I don't know the details on that.
Another factor to my risk profile is that one of the chemo agents I was given also has some maybe correlation to higher AD risk, though the research looks equivocal. I don't know if you know anything about that?
The other thing I worry about is that the aromatase inhibitor I have to take makes sleep less good. I take things to make sleep good, but the need for amyloid clearance at night feels important.
Up until recently I was someone who barely took Tylenol and had no health issues, and I avoided basically all medical care. It's been wild to enter into this world of extensive medical intervention and now to be kind of enthusiastic about it (not about getting intervened on, but about what's possible at the newer edges of what's available).
If you’re worried about it, it would be good advice to get enough sleep, and regularly. Stage 3 sleep seems to be the time when vascular fluid “washes out” the brain, potentially removing proteins before they become aggregations. Also it’s good to get sleep in general, so you won’t be wasting your time even if it doesn’t end up mattering for Alzheimer’s
> High intelligence seems to be a protective factor (hopefully this is reassuring to a lot of people here)
I don't think this is actually true. Rather, I think high intelligence is characterized as "protective" because, after suffering a lot of loss-of-function, formerly high-intelligence people still look normal to low-intelligence people.
But often, high intelligence people engage in more intellectual activity, much like physically gifted people are more frequently athletes. This continued intellectual activity late in life may be ,even if not "technically protective," still effectively protective by continuing to build new function, even as old function degrades.
Is there evidence that there is causation running from these to lower Alzheimer's risk? Assuming there's actually a correlation (Idk if there is), for several of these it sounds very plausible that the direction of the causation is that early Alzheimer's causes old people to stop maintaining strong social connections etc. unless there have been studies to disambiguate this from the direction of causation you claim.
I'm not sure whether people are just not allowing enough time for stage 3 sleep, or if people at risk for Alzheimer's don't have a good capacity for stage 3 sleep.
Thanks! That was a great overview and clarified a lot of aspects for me. In particular, I didn't know that the tau folding is so specific to its cause.
My background is that I'm a working neuroscientist in a field far from Alzheimer's research. From talking with colleagues, I'd often gotten the impression that amyloid was a dead end. So it was helpful for me to read the section titled "Challenges translating from mouse models." With that context, it's clear to me that a lot of the pessimism among neuroscientists stems from stories of those early, poorly-conceived mouse experiments.
Neuroscience research is often plagued with an unhealthy optimism that if you squint hard enough at an experiment in a mouse, it will seamlessly provide a good model for what's happening in humans. There's a large contingent of scientists ready to jump on subfields that produce large quantities of thoughtless work that suck up attention and available funds. This was clearly appropriate for some Alzheimer's research, but not for the field writ large.
Yes. IIUC, the side effects of the existing (approved) treatments for Alzheimer's are sufficiently bad that you're just as well off not taking them. Which is a pretty thorough condemnation.
A quick summary: The much-touted “27–33% slower decline” with lecanemab and donanemab masks the reality that these effects are clinically trivial, well below the minimum change patients or caregivers can perceive. Even donepezil, widely regarded as only modestly effective, outperforms these numbers. Trials last just 18 months, leaving it unknown whether these gains (imperceivable to caregivers) persist.
Meanwhile, risks are substantial. Amyloid-targeting antibodies cause brain swelling, microhemorrhages, and deaths, with the greatest hazard in APOE4 carriers and women. The APOE4 risk association requires some unpacking to be sure! See here: https://www.eneuro.org/content/11/7/ENEURO.0319-23.2024
But proponents of treating earlier in the disease course have already tested this hypothesis to disappointing results: The ambitious A4 trial tested this idea by treating older adults who had elevated amyloid but no cognitive symptoms. For 4.5 years, participants received solanezumab, an antibody targeting amyloid. Participants treated with solanezumab saw no cognitive benefit whatsoever—in fact, they performed slightly worse than those who received placebo. See here: https://www.nejm.org/doi/full/10.1056/NEJMoa2305032
Through a different mechanism in the late 1990s, this hypothesis was also tested with the AN-1792 vaccine that cleared plaques but did not halt progression to severe dementia. The trial was halted early due to severe brain inflammation and post-mortem examinations revealed a profound disconnect: despite dramatically reduced amyloid plaques, these patients had continued to develop severe dementia anyway. See here: https://academic.oup.com/brain/article/142/7/2113/5510133?login=false
The field’s response to repeated underperformance has been to shift diagnostic definitions and adopt weaker outcome measures (“time spent in mild stage”) that exaggerate benefit while obscuring minimal impact on real-world function. See here: https://www.nature.com/articles/d41586-024-00756-8
Many of these points are addressed in the article, but to respond directly:
Unlike donepezil, which only mitigates symptoms and doesn't affect the underlying biology (if you go off it, you end up where you would have been had you never been on it), the benefit from the partially-effective amyloid antibodies seems to get larger over time (open-label extensions point to widening benefits through four years), and persists vs. placebo even if you go off them. Despite this, they're not where we'd like them to be on either safety or efficacy, and I offer my belief as to why in the article: difficulty crossing the blood-brain barrier and treatment too late to prevent tau pathology. Thankfully both of these matters are being addressed.
I agree the side-effects can be serious, and are especially common for ApoE4 carriers. New BBB-crossing approaches like trontinemab seem to mitigate that risk greatly for everyone. A new donanemab titration schedule also appears to bring the ApoE4 carrier risk much closer to that of non-carriers: https://www.alzforum.org/news/conference-coverage/donanemab-small-tweak-titration-big-gain-safety
The solanezemab trial failed but this is one of the antibodies that didn't clear out plaques. In the study you link: "Amyloid levels on brain PET increased by a mean of 11.6 centiloids in the solanezumab group and 19.3 centiloids in the placebo group." (So a small between-group difference, but still a net increase. Whereas lecanemab, donanemab, and especially trontinemab cause deep plaque clearance.) The preclinical prevention hypothesis will be better tested by the clinical trials I mention in the post.
I had to read up on the AN1792 vaccine because I wasn't familiar with it. The main things I'd point out from that study are:
- The trial was interrupted because of meningoencephalitis in 6% of patients. [1]
- 5 of the 22 they examined post-mortem appeared to never have had amyloid plaques in the first place (sadly, it was uncommon back then to confirm Alzheimer's disease with biomarkers before clinical trial enrollment, which greatly reduced statistical power on many trials).
- They report that 7 of the remaining 17 only had "very limited" or "no" plaque removal.
- That leaves only 10 of 22 which had what they call "intermediate" or "very extensive" plaque removal.
- Treatment began after symptom onset, which is expected to be at best only somewhat effective under the ATN model.
- Despite this, there were hints of efficacy: p = 0.02 on their neuropsychological test battery for antibody responders. [1]
I wouldn't pin my belief on the amyloid hypothesis on this trial, but nor does it seem like clear evidence against it.
The anti-amyloid drugs we have now clear plaques more than enough to test the idea that removing amyloid will change the course of Alzheimer’s. That idea has been tested for decades, and it hasn’t delivered.
When you strip away the open-label extensions, which can’t prove durability because there’s no control arm and plenty of selective drop-out, the blinded trial data are all we can anchor to. The gains are statistically significant but far smaller than the minimal clinically important difference threshold, the point where patients or caregivers can actually feel a change. Even donepezil still compares at least as well in practical terms. Until a blinded trial shows a sustained, perceptible benefit, claims of “widening” or “persisting” effects are unwarranted.
Trontinemab may eventually offer a better safety profile, but right now that’s based on cross-trial comparisons and early-phase data. There’s already been at least one hemorrhagic death. We need fully powered, placebo-controlled Phase 3 trials before declaring the risk “greatly mitigated.”
And it’s worth remembering: even when AN-1792 produced substantial plaque clearance in half the treated group, those patients still progressed to severe dementia. That’s the core problem: namely, clearing plaques has not, so far, translated into meaningful change in the disease trajectory.
Instead of confronting that, the field keeps moving the goalposts. The latest proposed criteria would drop neurodegeneration entirely from the definition of Alzheimer’s and rely only on amyloid and tau biomarkers whose causal role remains unproven. Under that framework, many cognitively normal, amyloid-positive older adults would be labeled as having Alzheimer’s, while a neuroprotective drug that preserved function without touching amyloid or tau could be branded a failure. Critics have called for definitions rooted in clinical assessment and direct evidence of neurodegeneration, with biomarkers treated as unproven surrogates until they’re linked to patient-felt benefit.
Isn't it worth being a bit skeptical as to why these new definitions are being pushed? I am not anti-pharma by any means but there is a awful lot of money at stake here and these definitions seem bonkers given that ~1/3 of clinically normal older adults show substantial amyloid plaque deposition yet never develop Alzheimer’s disease during life. Keeping neurodegeneration as part of the criteria seems essential in distinguishing between normal aging and disease.)See a review by some alarmed neurologists here: https://pubmed.ncbi.nlm.nih.gov/38104639/
Until a treatment can meet that bar, durable, meaningful improvement at acceptable risk, changing definitions or endpoints won’t change what matters to patients and families.
FWIW, I mostly don't think you're addressing the points in the article or in my first reply. But I'll reply once more:
- First of all, my position is not "the existing amyloid antibodies are great!" It's "the amyloid hypothesis looks right, but there's still a lot of work to do".
- Second, I want to reiterate that treatment before symptom onset seems crucial under the ATN model, which has a lot of evidence behind it (discussed in the article). So the eventual progression to severe dementia after amyloid clearance is consistent with the ATN model.
- Third, I went into this investigation skeptical of the amyloid hypothesis, as you propose one should be. I was very prepared to find it lacking. I came away persuaded.
- Fourth, while OLEs do suffer from lack of placebo, the selective drop-out problem also applies to the ADNI and BioFinder controls they compared against. There's not an obvious reason to expect its effect to be much greater for lecanemab/donanemab than those groups. That said, of course we'd like to see longer comparison just against placebo in an RCT, and I certainly would concede that evidence from two OLEs is not by itself conclusive.
- Finally, in the absence of such long-term placebo-controlled trials, your position has less evidence behind it than mine. You're claiming we know that the drugs provide a clinically meaningless benefit because you're *assuming* that the effect does *not* widen, even though we 1) know the effect is disease-modifying, not merely symptomatic (since when you go off the drugs the effect persists, unlike with donepezil), and 2) have OLE data which at least offers some evidence, even if it isn't conclusive. Indeed, even if the drugs were 100% effective at slowing progression in those 18 months, the benefits would still be very dubious by your logic, since you're assuming (without empirical backing) that they won't continue.
... just for those slower ones in the crowd, ... um, what do you think about the ATN model?
it seems to me that arguing about how some idiots want to change the definition is not strictly relevant, no?
or, your last comment implies that we simply don't have enough high quality data to really test the ATN model?
because to disprove it we would need to get increasing-A and no-T (and no-N and no-AD) people, put them on A-clearing stuff, and wait, *and* if they still get T *and* N (and AD) then it was not the A?
or you are saying that the fact that millions of people have a lot of A and no AD (and so likely no N) means that the whole model is poppycock?
First, on legitimate criticism of amyloid primacy:
The skepticism isn't coming only from "no-nothing outsiders" but from prominent neuroscientists and drug discovery experts. Derek Lowe, a respected medicinal chemist with decades in pharma, has chronicled the field's endless moving of goalposts. As he notes, it's always: "We haven't dosed early enough, in the right patients, in the right way, targeting the right sort of amyloid." This refrain has continued for 30+ years while clinical benefits remain imperceptible to patients and caregivers. https://www.science.org/content/blog-post/had-enough-eh
For necessity, the definition now tautologically requires amyloid, but Primary Age-Related Tauopathy (PART) presents a challenge as these patients have tau tangles and can develop dementia without significant amyloid. The field debates whether this "counts" as AD, which reveals the circular reasoning problem.
For sufficiency, approximately 1/3 of adults over 70 have substantial amyloid burden yet remain cognitively normal. I’m not quite sure on the author’s position here but seems to argue these people would "eventually" get AD if they lived long enough: an unfalsifiable claim since many die cognitively intact at advanced ages. This breaks the sufficiency assumption.
Why definitional changes matter:
The proposed criteria would diagnose AD based solely on biomarkers (amyloid/tau) while dropping neurodegeneration and cognitive assessment. This has massive implications:
1. Medicalizes aging: One-third of cognitively normal elderly would be labeled as having "Alzheimer's"
2. Changes trial endpoints: Success becomes reducing biomarkers, not improving function
3. Creates treatment imperatives: Millions of healthy people become candidates for as of now risky, expensive drugs
As concerned neurologists note in the link I gave above: "Everyone agrees that AD is a form of cognitive decline caused by neurodegeneration, but neither cognitive decline nor neurodegeneration are tested for or given diagnostic value in the revised ATN system."
The public health nightmare in my view:
If we accept that treatment must begin at first amyloid detection (before tau), we're proposing decades of prophylactic treatment with:
• Unknown long-term consequences (antibodies are notoriously promiscuous in binding; what are the off target effects over years?; antibody therapeutics can be especially problematic in people with autoimmune risk, so what additional harms could we be causing?)
• Enormous costs ($26,000-56,000 annually per person)
• No way to distinguish who will progress vs. remain healthy
Alternative frameworks:
What if amyloid accumulation simply reflects impaired protein clearance with genetic susceptibility overwhelming the equilibrium decades earlier? The glymphatic system's role in brain clearance has only recently been appreciated. Sleep disruption, vascular changes, and inflammation all converge on clearance pathways. This suggests targeting clearance mechanisms rather than treating everyone with detectable amyloid for decades. This is just one alternative therapeutic target that can assemble the evidence into narrative cohesion, there are many more to be plumbed.
The ATN model may capture one pathway to neurodegeneration in susceptible individuals but treating it as universally deterministic ignores the millions who age successfully despite amyloid. The push to redefine AD around biomarkers rather than clinical syndrome serves commercial interests more than patients. All we can do now is return in another decade to see whether the next iteration of "earlier, better, more" finally delivers or whether we're still hearing the same refrains.
It's neither. It's rocket *plumbing* that's hard. If the plumbing works, the science and engineering are generally A: straightforward and B: fun even when they get tricky. But rocketry is pretty much a worst-case environment for plumbing nightmares.
I still think it's probably a mistake for drug research to target amyloid. Amyloid is produced because it has many benefits. So it seems somewhat likely that the high amyloid levels are a mostly desirable reaction to some other problems, and more attention should be devoted to the problems that are upstream of amyloid.
If this were the case, we'd expect ApoE3 homozygotes — and others who are not at much higher risk of Alzheimer's due to increased amyloid production or impaired clearance — to be dropping dead of infection left and right, or at least frequently enough to outweigh the much reduced risk of Alzheimer's. But we don't see that.
As a former (systems, not molecular) neuroscientist, I don't find this terribly convincing? Claims like "gene X does only Y" have a very very poor track record, and brains of a certain genotype may have perfectly good reasons for what seem to us like inexplicably excessive amyloid levels. This is also consistent with reports of negative side-effects from existing drugs.
Having said that, I think we must keep trying on amyloids and dealing with the consequences as they come.
It’s not so much “gene X does only Y”, but more “Y is the only plausible mechanism that all the risk genes have in common, and for some of them the only plausible mechanism that has been proposed".
That's a very reasonable claim, but I don't think it is a sufficient answer to Peter's objection.
I think you can answer Peter by saying "well, Alzheimer's is very bad, so any problems we cause by clearing amyloids are likely to be less bad than Alzheimer's, and anyway over time we can minimize them with calibration/optimization/compensation".
I don't see how you can get a clear prediction about Apoe3 people dying from infection without making some assumptions that I consider to be questionable.
For example, some of the reduced amyloid with Apoe3 could be the result of Apoe3 handling pathogens better, so that infections less frequently reach the stage where amyloid is called for. If that effect is important, I'd expect Apoe3 people to die from infections less.
Apoe variants have important effects on infectious diseases. I suspect that amyloid-related differences only cause a small fraction of these effects.
I also note that there are diseases for which Apoe3 is clearly worse. So your observation that we don't see Apoe3 people dying more is fairly dependent on where you look.
The more I look, the more complex this topic seems.
OK, so I understand your view to be that ApoE4 carriers are worse at fighting off infection, and this means that infections in ApoE4 carriers are more likely to reach the point where the brain resorts to amyloid as a defense mechanism, and that while this defense mechanism then leads to Alzheimer’s in many cases (~80% lifetime risk for homozygotes), it’s still worth it compared to the alternative, which is a very bad infection. Is that right?
If so, a few points about this:
1. Alzheimer’s is so bad that the infections in question would have to be extremely bad or extremely common to justify that tradeoff. While this is possible, I’m not aware of any strong reason to believe this would probably be the case.
2. This seems to predict that those treated with aducanumab, lecanemab, donanemab, and trontinemab, all of which clear out amyloid plaques pretty well, would get many more and much worse infections than those treated with placebo. But there is no hint of this in any of the safety data: the main frequent and serious adverse events are brain bleeds and swelling related to the unintended congregation in blood vessels; and trontinemab, which mostly avoids this, has much less bleeding and swelling while achieving much better amyloid clearance.
3. You’re not eliminating all amyloid, only the fibrillar plaques, and only in people who are very likely to progress to dementia without treatment because they already have an amyloid burden which would cause that. This means the antimicrobial properties of soluble amyloid species, if they are very important, can still mostly function (just as they managed to function for the first ~50 years of the person’s life without severe amyloid plaque burden). And once you clear out the plaques, treatment might not be necessary again for years (amyloid plaque burden grows pretty slowly, including after clearance with antibodies).
It's quite plausible that overproduction of ANY of the body's proteins/hormone/etc. can be hazardous. It's what you should expect. And you should also expect every system to be interconnected with lots of other systems in ways that aren't obvious.
I truly expect that if you could clear out the (excess?) amyloid plaques without doing other damage it would be beneficial. Also that if you could eliminate the tau misfolding it would be beneficial. And it's quite plausible that these are connected. I also have a suspicion that the tau misfolding is more directly causal of Alzheimer's. If the amyloid causes tau misfolding, then removing the plaques should not be expected to solve the problem after it has initialed tau misfolding, because I believe that they tau misfolding, once initiated, is a prion disease. But it might well slow it down by halting the creation of new roots of misfolding trees.
Aducanumab, Lecanemab, Donanemab, oh my. Since they all end in ....mab I'm assuming there is some method to drug naming, does anyone have a pointer to the decoder ring? I had always just assume pharmaceutical companies had terrible marketing departments.
-mab means it is an antibody therapy, -ib is a small molecule inhibitor (typically kinase inhibitors). -vir are antivirals. More naming conventions: https://en.wikipedia.org/wiki/Drug_nomenclature
you get used to it. except with cell and gene therapies — a lot of their official generic names are so hideous no one even attempts it and short nicknames are universally used instead. i once came across a conlang someone had constructed wholly out of drug nomenclature, and spent an afternoon both 1. elated by it and 2. writhing in frustration, because none of the lingustic design choices actually corresponded to or were meaningfully rooted in the drug nomenclature? i just really wanted there to be, idk, assignments of such and such a drug class/suffix to such and such a bit of grammar or syntax that *made sense* lolsob
These are scientific names, like "acetaminophen", not commercial names like "tylenol". Commercial names are the ones that sound like they come out of marketing departments... most likely because they do.
Funny you use that Tylenol as an example, got me interested so I had to look it up on wiki: "Like the words paracetamol and acetaminophen, the brand name Tylenol is derived from a chemical name for the compound, N-acetyl-para-aminophenol" I can't figure out how to add bold font here but wiki has bold for the tyl at the end of acetyl and enol from the last four letters of aminophenol.
Reading those always makes me wonder if they are actually reversed names. Itd be funny if their eventual brand name is indeed its clinical name reversed.
I guess my question is: if this is a disease caused by amyloid overproduction, and we know which genes cause amyloid overproduction, can't we just do a CRISPR thing and, like... turn those genes off? Turn them down a bit?
Note that *most* instances of the disease are caused by impaired clearance, not overproduction. However, it is still possible that turning down production can help restore the balance.
Can some of the actual neuroscientists and medical practitioners chime in about having "smart guy from SpaceX" enter into the medical field of deciding what constitutes promising research after doing six months of reading?
This was a great piece and it seems convincing, but I don't know the field.
The last time a bunch of smart guys from SpaceX got into medical research.... It destroyed the NIH. I know some people forgot this ancient history because it was over a month ago, but a little bit of "I'm sorry that some of my former coworkers at the behest of my former boss cancelled funding for vast amounts of promising therapies in all medical fields, but here is my piece."
The idea that David Schneider-Joseph has worked at SpaceX in the past seems like interesting background but not relevant to his post. If he hadn't worked at SpaceX but had written the same words would that make his post more compelling?
Part of what we look for in anyone is "do we trust their background and who they are". We licence physicians and lawyers as a way to tell non specialists that the licensed individuals have a certain degree of training.
OK, so him disclosing not being a physician or neuroscientist is good.
If his previous job was being a plumber, that would actually be better.
But his previous job was SpaceX. Elon Musk's recent conduct at DOGE was to empower some very bright people (in his estimation) to take a blowtorch to medical grants that literally tens of thousands of researchers and millions of patients were depending on.
Why? Because Elon Musk is a smart guy. He and his underlings could decide what was good medical research... Trust them, they are smart and many of them worked on rockets!
So reading this, I am seeing "former SpaceX engineer says he knows better than medical professionals." That is a super negative prior for me personally based on what other people with similar biographies just did to American medical research.
I'm sorry that some of my former coworkers at the behest of my former boss cancelled funding for vast amounts of promising therapies in all medical fields, but here is my piece.
(In all seriousness, I'm very upset with what DOGE did.)
I write this not to tell Alzheimer researchers what they should be doing, but to convey what I learned about the field to other people, like myself, who are not in it.
It might be that you are not a lifelong medical researcher that contributes to the high readability of this piece. You've obviously learned an enormous amount about this issue in six months, and maybe it's your ability to think like someone not embedded deeply in the research that helps you translate what you've learned into such a readable piece for the rest of us.
> I am seeing "former SpaceX engineer says he knows better than medical professionals."
I know the author has already replied, but I want to point out I think there's a serious basic flaw with this observation.
The Amyloid hypothesis is still the orthodox position in the field. It's more contrarians and outsiders who have claimed that it's false. So my read is that this post is more "here's why the experts ARE right.
But actually, more basically: When there is disagreements amongst experts, you can't agree with *both* sides. So that means EVERYONE who is convinced by one side or the other will find themselves believing they know better than (some) experts; it's unavoidable, unless you're willing to hold off on making conclusions until and unless there is consensus in the field.
What he's actually saying is more along the lines of "hey, it looks like the experts were on the right track all along, despite the superficially believable arguments given by contrarians. This is a hard problem, we shouldn't abandon what they're doing just because it hasn't yet had positive results yet."
Man I wish the people at Doge were this thoughtful.
Well, Scott posted this, so presumably that's at least a once-over from him.
It's true that skepticism is warranted when people Do Their Own Research™ outside of their area of expertise, but for really high quality people it's not impossible for it to turn out well. I mean, what, do you want make it taboo for non-doctors to engage in any sort of public discussion of medical research? All that's necessary is for the person to readily disclose that they don't have formal training, which he did, and for the audience to keep that in mind while reading.
>a little bit of "I'm sorry that some of my former coworkers at the behest of my former boss cancelled funding for vast amounts of promising therapies in all medical fields, but here is my piece."
That level of guilt by association is not remotely healthy. He has no obligation whatsoever to preface an article completely unrelated to Elon Musk with something like that. You are trying to bully someone into going out of their way to help you spread your political opinions. Cut that out - in general, but especially in ACX comments.
Former neuroscientist: this seems fine? Specialists will have their careers and egos and grants tired up with very narrow sub-problems in Alzheimer's research---in this case, for instance, a particular dead-end mouse model---so a disinterested outsider may have an easier time seeing the big picture. Just obviously don't take it as gospel.
In terms of the substance of this post, the main red flag I noticed was this statement: "The clearest relationship is various genes which massively increase amyloid production (while doing nothing else)". In neuroscience, claims of this sort are almost always wrong and must be viewed with intense suspicion.
I agree it may be close enough to being true for the problem at hand, and that specific claim wasn't especially load-bearing for the rest of the post anyway, which does a nice job laying out the convergent evidence. I probably should have called it "the only yellow flag" rather than "the main red flag" because it really was a very good post.
Responding late to your edit, but my point wasn't that APP and presenilin or any other biomolecule have other specific known functions; it's that in biology, you will never go broke betting that there is another function.
The practical implications here are... not much really. The amyloid hypothesis laid out here sure sounds like the best bet available. I'd just downgrade the probability and/or expected degree of success.
Good to hear it passed the smell test. As a non-scientist who has worked in drug and device development, it struck me as lacking epistemic humility. The track record of non-specialists wading into biology is not great, often because there’s a tendency to a) over-index on a single hypothesis and b) over-index on hypotheses that can be understood by a non-specialist. It would be much more credible if it covered all major hypotheses being explored and the evidence for or against them.
I am not a neuroscientist but I am a biomedical researcher who has followed Alzheimer's research for years -I think David did a good job summarizing the research consensus in a readable fashion, and I have no problem with this in general. Often researchers deep in the weeds have trouble explaining their field in a way that is easy for outsiders to understand, and they often are too busy with research to do even if they have the ability. In my own field of virology, I think Peter Miller did a better job than any virologist so far (including myself) of explaining the evidence for and against the lab leak hypothesis, and I've seen many colleagues with the same sentiments.
The real tricky stuff comes when an outsider with no relevant expertise tries to *disprove* a widely accepted hypothesis, instead of summarize a scientific consensus...still not out of bounds, but the odds of success are...much lower.
> The real tricky stuff comes when an outsider with no relevant expertise tries to *disprove* a widely accepted hypothesis, instead of summarize a scientific consensus...still not out of bounds, but the odds of success are...much lower.
Nice heuristic; I'm planning to still this summarization.
>Can some of the actual neuroscientists and medical practitioners chime in about having "smart guy from SpaceX" enter into the medical field of deciding what constitutes promising research after doing six months of reading?
I agree with this general sentiment. SpaceX/DOGE controversies aside, six months reading literature in your spare time is not nearly enough time to really understand a large, sprawling, comprehensive field, no matter how bright you are. You wouldn't ask for an authoritative review of a field from a first-year PhD student, and they aren't even working another job full-time!
I don't want to detract from this piece, nor David's abilities -- it's very well-written and rigorously researched. And I don't want this to turn into "Trust The Experts (TM)"; I think generally it's fine / useful for smart and capable people to write deep dives on some topic they get interested in, but this is exactly the kind of topic where a true credentialed expert in the field (not Scott, not David, not me) is the only person who could persuasively write "the defense of the establishment view on amyloid hypothesis." I have just seen too many examples in my own field of biomedical research where smart outsiders get one-shotted by first-year PhD-level pitfalls to fully trust something not written by a credentialed expert.
That David should apologize for the actions of some of his ex-colleagues working in totally unrelated roles seems wild to me. He made a wonderful effort post here that's very well footnoted and seems like a gift to all of us.
Did he not site enough papers for you? Are you looking to be certain he is correct without bothering to read any of the dozens of references? Why not just make a Bayesian update to your priors and move on....
You mention "extremely rare protective genes" that can protect some people even after the amyloids take hold. Has there been any research into trying to copy those genes' effects for a later stage treatment?
Lexeo is trying to deliver protective ApoE variants via an AAV vector, and in some drugs (LX1020) also suppressing expression of the pre-existing harmful variants: https://www.lexeotx.com/programs/cns-programs/
I think the evidence is pointing to the harm (at the tau stage) mostly coming from ApoE4's gain of toxic function rather than loss of protective function, so I'm somewhat optimistic about LX1020 but not as much their others. But it's early for them.
It seems like somatic gene-editing could kill a lot of birds with one stone here, but I don't have a great idea about how safe/cheap/reliable the delivery vectors would be at the moment.
Yes, but that "somatic gene editing" would need to be done on the glia, or perhaps neurons, within the brain. This is both difficult and dangerous. Also, I believe that tau misfolding is a prion disease, so once it happens, fixing the genes won't fix the problem (though it might slow it down a lot).
My guess would be that's a problem at least as hard as "design a new laundry detergent to re-weave frayed fabrics," or "breed a new type of tree to proactively extinguish wildfires."
Could I suggest adding a little summary at the end of the A->T->N pathway, and, in particular describing your argument that it's expected that anti-amyloid drugs have failed, given how they're used so far? (i.e. only after neurodegeneration has started). That was a surprisingly momentous/hopeful (and plausible) claim for being kind of just buried in the middle: that we basically already have a cure, and all that's needed to make it work is to scan everyone for amyloid buildup, so they can get the treatment a decade before symptoms would have started.
These extremely technical medical posts on here usually get the skimming treatment from me (and I suspect from a lot of other not-medically-literate readers) so ending summaries can be really helpful to reaching a wider audience! (I know you basically already have this under "The mechanistic claims", but that's still buried halfway through).
Also... I'm surprised this argument hasn't been put out in the press more. The total failure of anti-amyloid drugs seems like pretty damning evidence against the hypothesis, and such a plausible mechanism simultaneously explaining that failure and promising a cure seems like it would be an eyeball magnet for journalists.
This is a dense but highly readable analysis of the problem and I was surprised I stuck with it to the end. One thing I am curious about, is if AI can be of any help at leading to the proposed 75% solution mentioned under the testable prediction heading.
Neuroscientist working in an unrelated area, but I wanted to mention that a discussion of AD is not complete without mentioning the calcium hypothesis.
There is growing evidence that calcium dysregulation can drive both both amyloid aggregation and Tau cleavage, so there's the possibility that it may be upstream to both. Sleep disruption and TBIs also directly impact calcium homeostasis, so there's a mechanistic link.
I want to join in and praise this excellent summary! I was also skeptical of the amyloid hypothesis from a much more superficial engagement with the field 10-15 years ago, and you have convinced me otherwise.
One thing that you have to learn in biology is ignoring outliers, and I think that was something that I got wrong.
You used the phrase "all instances" in your prediction, which is fine as rhetoric device, but is not quite how reality works. *Almost* all patients with a clinical Alzheimer diagnosis have amyloid plaques revealed by PET. Where "almost all" apparently means 88%. I think this made me severely suspicious when I was younger, because it means that 12% of the people with Alzheimer apparently did not get it via amyloid. But I have learned since then that in biology, 88% means "almost all", and the rest is a lizardmen fraction where something was weird. Perhaps the PET measurement wasn't accurate, perhaps the diagnosis was wrong, perhaps the patients did have some sort of exceptional Alzheimer which developed via some weird different pathway.
But it does make me skeptical of your prediction. Not about the direction, but 75% slowdown seems huge. Because not 100% of patients will react to the therapy, some will not respond for unfathomable reasons. And if 25% don't respond, you would have to slow down the disease by 100% in the rest, or more realistically, the therapy would have to revert symptoms in some patients. Which may happen, but which seems a strong prediction even in your bullish view.
It's true that in the past, Alzheimer's disease was diagnosed without biomarkers. Nowadays, the presence of amyloid pathology is considered part of the diagnostic criteria.
This doesn't make my mechanistic claim merely definitional or vacuous, because I framed it as a *counterfactual*: I'm saying that in all instances of Alzheimer's dementia (which I define in the following sentence as a whole collection of things including amyloid pathology), then counterfactually, amyloid-only therapy *would have* prevented dementia. (Of course, we can't directly test a counterfactual in individual cases, but that's why I made a separate prediction that can be tested.)
In an earlier draft, I had the parenthetical "(barring occasional instances of some other, separate and clearly distinct disease process, such as Parkinson’s disease dementia or a tauopathy with a different fold and different pattern of regional progression)". It was removed for brevity, but I still endorse it.
Ah ok, my number obviously doesn't make sense anymore if amyloid pathology has entered the diagnostic criteria. And thanks again! I still find the 75% in the testable prediction extremely optimistic, but it's not a fundamental disagreement. And time will resolve what the numbers are in the end.
This is a statement not a question: obviously a common reason for a non-specialist to get very interested in a medical disorder is because they or a loved one suffer from it (or are likely to). I hope that’s not the case here; if it is, my best wishes.
> Severe enough amyloid pathology is a sufficient cause of Alzheimer dementia in almost all brains.
I haven't seriously investigated this issue, but I thought this was ruled out by the nun study? Or, like, they came up with the idea of 'cognitive reserve' that meant you could have lots of amyloid in your brain and not have Alzheimer's because you were losing brain you weren't using and still had enough left, but I thought it was pretty suspicious that the first time they looked at a significant number of controls the relationship got substantially weaker.
I discuss the existence of amyloid-positive, cognitively normal individuals in the article. The nun study (which I did not specifically mention) is one frequently-cited study which shows the existence of such people. This is predicted by the ATN model since it says amyloid does not directly cause neurodegeneration, but does so mediated by tau, after a ~15 year preclinical phase (I detail all the evidence for this model in the article). The question isn't whether such people exist, the question is whether even the ones with extremely high amyloid burden (not merely nominal positivity, since there are varying levels of susceptibility to the A->T conversion) ever go ~30 years without dementia.
In a PET study (https://academic.oup.com/jnen/article-abstract/70/10/832/2917208 figure 4), you can see that some cognitively normal people with "amyloid positivity" (> 23 centiloids) did not develop tau pathology within five years, but nearly all of those people were below 60 centiloids, and none of them were above 100 centiloids, which is approximately where I might draw the "sufficient amyloid" line. The claim isn't "all amyloid positivity results in Alzheimer's dementia", it's that "amyloid negativity prevents Alzheimer's dementia, *sufficient* amyloid burden basically guarantees it, and in between you have monotonically increasing risk".
(Overall, I don't find "cognitive reserve" to be a very useful concept: I think it's more an artifact of dementia and mild cognitive impairment diagnoses employing cutoffs on cognitive tests, not any actual sparing from an underlying disease process.)
Very well written and interesting post. I've seen my father drift off in his eighties. He did much much more headers than me, maybe I got knocked out once or twice more and I like my drink maybe even a bit more than him. I think it will be ok when my time of dementia will arrive. I don't have that much to lose.
Why is it easier/more common to target amyloid than tau? Amyloid seems worse since it's not the proximate cause of neuropathy and needs to be targeted or symptomatically.
Also, is there useful work on identifying the early amyloid plaques cheaply so we can target anti-amyloid drugs when they're most useful?
I think it’s mainly just that the importance of tau in mediating the A->N connection only became clear about 20 years after the causal evidence for amyloid itself, and drug development takes ~15 years from the first preclinical experiments, to early failures (such as N terminal tau antibodies), to trying more promising approaches (such as midregion antibodies and the IMO even more promising antisense oligonucleotides).
As for identifying amyloid pathology early: yes, this is possible to do today (via PET, CSF, blood, or plasma) but it’s not common because there haven’t yet been proven prevention methods. But once prophylactics are proven, I assume it’ll become widespread practice.
Very interesting article, thank you. I notice that I think it makes me a bit more confident than the amyloid hypothesis is correct compare to the other but on the meta level it makes me feel like a weathervane.
One question: I lost a grandparent to Alzheimer's. I am now worried about one of my parents. If I understood everything correctly we would want to start treatment something like 15-20 years before Alzheimer's start striking, which if the timeline are the same for my grandparent and my parent should be in the next decade or so. What can I do? What is possible to do?
I remember a period after they found fraud in research supporting the role of soluble amyloid beta. There was a social media storm, and as far as I could tell nobody participating in it understood that this was about soluble amyloid beta, which is actually a different hypothesis than a role for amyloid beta plaques. Your source [123] seems to back up my memory. The subtitle is "A neuroscience image sleuth finds signs of fabrication in scores of Alzheimer’s articles, threatening a reigning theory of the disease", but the "theory of disease" seems to be a causal role for "toxic oligomers". Was this really "a reigning theory"? Can't there only be one reigning theory? And I would have thought that's plaques. Anyway, I think a lot of people casually following the story got mixed up.
There are many species of amyloid beta, varying in length, aggregation status, solubility, fold, and other post-translational modifications. Even though the causal role of amyloid has been apparent for a while now, there still hasn’t been a ton of clarity until possibly relatively recently about which species are “the toxic ones”, which ~20 years ago, before the mediating role of tau was as well understood, mainly meant directly causing neurodegeneration. Some preclinical evidence (including some of the fraudulent work) pointed to certain oligomeric species, but this was never a consensus, and was never the main reason people believed in the amyloid hypothesis.
Now we’re understanding 1) that the toxic effect is mostly mediated by tau, so we should be mainly looking for amyloid/tau interactions rather than direct neurotoxicity, and 2) that the harmful species are in greater states of aggregation and less soluble, such as plaques or protofibrils.
I would bet on the following: the leading* therapy in 12 years will not be one whose sole intended mechanism involves amyloid production or clearance, e.g. monoclonal antibodies, small molecules, or gene editing that work directly on amyloid.
“Leading” means some combination of efficacious and most widely used.
This is the important sense in which the amyloid hypothesis is wrong: it may be “correct” in a narrow biochemical sense, but it is not helpful.
When did those scientists realize that they need to treat amyloid BEFORE it's too late? Recently? Otherwise why were there those late treatment experiments? I guess we do realize it recently, maybe only because of those failed late treatments?
I think the consensus gradually emerged over the last decade, mainly due to:
1) The increasingly clear importance of intermediate pathologies like tau which mediate the effects of amyloid.
2) The modest clinical benefit of drugs like lecanemab, and donanemab, despite significant plaque clearance.
A challenge when conducting clinical trials is you ideally want to test in a population which is likely to decline in the course of the trial, and for the trial to not run too long, so as to maximize statistical power, minimize cost, and maximize speed of discovery. So until it became very clear that earlier treatment was vital for amyloid therapies, there were natural pressures pushing in the opposite direction.
Question about the clinical trials of lecanemab, donanemab, aducanumab. I understood that the clinical trials calculated the change in the speed of progression by looking calculating (cognitive score at T2 - cognitive sore at T1) / (time between T1 and T2). This doesn't rule out the effect being to simply improve cognition by a fixed amount, without impacting progression. Is my reading correct, or did I miss something? Asked differently, is the improvement from these drugs different in kind from the improvement from memantine or donepezil?
An important difference compared to donepezil (and other acetylcholinesterase inhibitors) and memantine is that 1) the advantage over placebo plateaus after a few months, and then your trajectory of decline matches placebo but vertically shifted and 2) when you go off those drugs, you revert to the trajectory of the placebo group.
I don't have any relevant expertise or knowledge, so the following probably just reflects my ignorance. But if we stipulate the ATN model as OP explains it, it seems to me that there are daunting problems with targeting amyloid (and so I don't understand why OP is so optimistic about it).
If treatment really has to start 10-20 years before symptoms to be fully effective (because tau degeneration is self sustaining once kicked off) then to significantly impact the disease (a) you would need to aggressively screen and treat relatively young and unsymptomatic people, and (b) proving that the benefits of doing so outweigh the risks will take decades. Is society really going to decide to start giving every 50 year old a PET scan and then putting some significant fraction of them on monoclonal antibody infusions for decades (that might have significant side effects) based on extrapolation from 18 month studies in much older, symptomatic patients that (can at most) show small, statistically signficant but clinically insignificant benefits? Is anyone even going to try to run such a difficult randomized trial over the course of 15-20 years, in hopes that the results will help people in 35-50 years? Treatment that attacked a later stage of the A->T->N pipeline would have a huge advantage in timeline for both testing and deployment.
Thankfully, there are plasma tests that have recently been validated and aren't too burdensome or expensive to administer.
As for results from prevention trials, we should get results for donanemab's in 2 years, and lecanemab and trontinemab in a small number of years after that. While that trial may not involve treatment quite as early as I'm suggesting, if the earlier = better principle is correct, we should hopefully start to see substantial improvements in efficacy compared to current approaches, and I'd imagine there'd be a shift towards treating as early as possible.
The main thing is that the first people who get prophylaxis or treatment in the asymptomatic phase of the pathology will be people with extremely high risk or even the AD or chromosomal forms of Alzheimer's. They will be eager for treatment and the risk/benefit will look better for them. Data from those people will inform management for patients with lower risk.
There isn't enough money for widespread treatment at $30k/dose; my expectation is that ASI/nanotech will soon reach the point where everyone is either cheaply cured and on UBI, or dead.
>This fraud has sometimes been framed by amyloid hypothesis critics as impacting very foundational work for the amyloid hypothesis. But this mostly isn’t true.
Excellent summary! After the fraudulent research was exposed, I wasn't sure where we stood with the Amyloid theory. Of course, science journalists are mostly ignorant of the subjects they write about, and reliance on popular articles will give you a distorted view of the questions. I never had the time or the motivation to dig into the literature. Thanks for doing what the science journalists should have been doing. :-)
I know nothing about this topic but I felt an overwhelming urge to come here to plug the book Valley of Forgetting by Jennie Erin Smith, about the lives of a group of Colombian families with autosomal dominant PSEN1 mutations (as mentioned in the post) and the scientists who have been studying them for the past few decades. It's a really good book. That's all.
Thank you for this. I am confused about your bet. If amyloid build up typically predates taopathis by 15yrs and amyloid alone doesnt cause much cognitive decline, why would you expect studies targetting amyloid build up to be completed within 12 yrs? They have already begun?
Some have begun; and also plausibly one only needs to clear amyloid pathology prior to significant tau pathology, not at the very beginning of amyloid positivity; and also my bet is only aiming for 75% efficacy (I’d expect more over time).
So far, our evidence comes from non-randomized studies of people with diabetes (comparing those on Ozempic to those on insulin). There are parallel phase 3 RCTs due to finish next month, in people with Alzheimer's disease, which will be a better test of the theory.
Is it just me, or is there a slight bait and switch going on here?
This is not a topic I'm that informed about. But my simple understanding of the amyloid hypothesis was indeed that amyloid is directly causing the symptoms, which considering all the early antibodies and mouse models seems to also have been what researchers though; Wikipedia still has the amyloid hypothesis as amyloid directly causes the symptoms without mediation. And the standard arguments against the amyloid hypothesis likewise focus on the bad correlation between amyloid plaques and impairment.
From a cursory search, it indeed looks like currently, amyloid hypothesis in the research is mostly used in the amyloid tau cascade model sense you describe. Which does sound more convincing than the original hypothesis. I don't know if this is simply a failure of communication, but it seems like what the amyloid hypothesis refers to has quietly changed in the last decade or two, without that percolating in the regular critical discourse about it. I'm undecided as to whether it has changed so much that it should really be considered a different hypothesis that should be given a new, distinct name. But clearly, the amyloid hypothesis as stated twenty years ago is false even according to your model of the disease.
I think this is fair! This seems analogous to how "atomic theory" turned out correct but it was later discovered that what had already been called "atoms" weren't the atomic units after all. Is this a bait and switch? Well, maybe! I don't know enough about the history there, but I imagine it was a combination of 1) some terms undergoing a natural scientific evolution as our understanding improved, 2) some scientists protecting their egos, and 3) the challenges of communicating science to the general public, who have a very low tolerance for complexity. I think there's probably a similar breakdown in this case.
I only mean to defend the ATN model, and more generally the idea that amyloid plays a necessary and (in sufficient severity) sufficient role, not the idea of proximate neurotoxicity.
Do you know which model was the impetus for the current antibodies in the pipeline? If amyloid is the ultimate cause it's probably a good target for prevention. But if the actual toxic effect goes through tau that would seem to be the obvious target to treat people post symptom onset (instead of decades beforehand). I'm not aware of any tau antibodies currently being tested, is it fair to assume that so far, all treatments that made it to clinical studies did indeed assume that amyloid is directly causing the symptoms?
To your first question, depends on how you define "in the pipeline". The amyloid antibodies currently on the market began development as far back as 2007, in particular lecanemab's initial murine form "mAb158" [1], motivated by the so-called "Arctic" Alzheimer's disease mutation, which was known to result in more amyloid-β protofibrils but not actually yield plaques. So the theory there is that it's the protofibrils (rather than the less-aggregated species like monomers and dimers) but not plaques *per se* which were the problem.
In that paper, they say that "oligomeric Aβ, including the protofibril, is thought to be a primary neurotoxic agent in AD", which was a commonly held view at the time. So that would indicate a belief in a fairly direct neurodegenerative effect.
Donanemab development began in 2012 in mice as "mE8" [2], motivated by a perceived need to target plaques rather than less aggregated, soluble Aβ (which is slightly different from but reconcilable with the lecanemab rationale). They suggest a more indirect viewpoint, saying "amyloid deposition is the causative factor for the initiation of the neurodegeneration cascade, which includes inflammation, gliosis, neuronal damage, synaptic loss, and cell loss", though they don't mention tau specifically.
By now, virtually everyone who believes in the amyloid hypothesis, including those working on e.g. trontinemab, accepts that most neurodegeneration is mediated by tau pathology (or something closely connected to tau pathology).
Most clinical trials since the year 2020 (but not before) have had targets other than amyloid (see my references [126, 127] in the essay), with tau probably the most common non-amyloid target. Four antibodies (semorinemab, gosuranemab, zagotenemab, and tilavonemab) failed, but we now have much better structural understanding from cryo-EM which suggests these antibodies were targeting the wrong part of the tau protein to prevent the prion-like misfolding.
Newer antibodies such as bepranemab (UCB0107), posdinemab (JNJ-63733657), etalanetug (E2814), and BMS-986446 target tau's midregion and have a better chance of success. bepranemab is the only to have completed phase 2 so far (posdinemab should be next, in about six months). It slowed tau accumulation by ~half but failed on its primary cognitive endpoint, although several secondary cognitive endpoints showed indications of benefit, and in a pre-specified subgroup analysis in a "low tau" population (so the very early symptomatic stage, when tau pathology is just getting started), this signal was a bit stronger.
One explanation for this may have to do with tau pathology worsening by two basic mechanisms: intercellular spread and intracellular replication. Antibodies mainly act extracellularly, which can help to slow spreading in the early stages, but once there has been enough spreading, most worsening is occurring intra-cellularly and requires a different strategy. This is like how early in a pandemic shutting down air travel is helpful, but once there's a little bit of virus in every region, local efforts are going to be necessary too.
A strategy consistent with this hypothesis is antisense therapy, such as BIIB080 (MAPTRx), which suppresses expression of the tau gene to provide a smaller "susceptible population" for the prion-like replication. In its phase 1b trial, it didn't just slow tau pathology — it actually *reduced* tau pathology, which is the first time this has been achieved in humans. And while not powered to detect cognitive benefits, it appears to have slowed actual cognitive decline by about 2/3 (with very wide error bars) vs. controls from a different study. So I think this is the most exciting approach for targeting tau. Phase 2 began last year.
So I guess I should withdraw my bait and switch accusation, although the phrasing is definitely confused in the literature. From a 2016 paper on aducanumab:
> The amyloid hypothesis posits that Aβ-related toxicity is the primary cause of synaptic dysfunction and subsequent neurodegeneration that underlies the progression characteristic of AD. [1]
Which does kind of sound like Aβ is directly killing your neurons to me. But then they reference a 2002 paper for that claim, which seems much more carefully phrased and aligns well with your argument, even explicitly mentioning tau:
> According to the amyloid hypothesis, accumulation of Aβ in the brain is the primary influence driving AD pathogenesis. The rest of the disease process, including formation of neurofibrillary tangles containing tau protein, is proposed to result from an imbalance between Aβ production and Aβ clearance. [2]
So maybe all the focus on Aβ antibodies just merged that in the mind of most non-domain experts talking about the topic (and maybe also some domain experts). Anyway, thank you for restoring some of my faith in medical reserach (or this particular field of it). Apparently part of the issue here was just the long time between initial development and clinical trials, and the field is indeed capable of eventually changing course once it becomes clear that the current approach isn't yielding any results.
> After it forms, the tau pathology no longer appears to require amyloid’s assistance to keep spreading (although amyloid may still accelerate it). This probably explains why existing anti-amyloid therapies have been only ∼30% effective in test patients, who are usually late in the amyloid → tau progression even if early in having symptomatic disease.
This would seem to suggest that anti-amyloid therapies _would_ work, if we gave them to people who were going to develop Alzheimer's disease in the future rather than giving them to people who have already developed Alzheimer's disease in the past.
The post suggests that many such people are easy to identify. What does this experiment show?
Overall, this is a truly excellent piece of exposition. Thank you. I think the "frustration" section is unnecessary and possibly detracts from the piece as a whole.
I'm worried that it often takes this level of 1) training/knowledge, and 2) willingness to spend 10s of hours on a single paper, to see the actual problems.
I think it's right to criticize the old highly over-expressed amyloid-only mouse model for not being a full recapitulation of the disease (doesn't have tau neurofibrillary tangles). But it:
1. Doesn't engage with all (or really any) of the extensive evidence I discuss, from amyloid+tau mouse models and humans, for amyloid → tau causation. In the human disease, tau pathology has an origin independent of amyloid (so isn't captured by the amyloid-only mouse model they focus on), yet amyloid is also necessary (and in enough severity) sufficient for that pathology to extend beyond the medial temporal lobe (MTL), and usually for it to locally worsen within the MTL.
2. Doesn't offer an alternative hypothesis to explain the genetic evidence for amyloid's causal role.
3. For antibodies which do clear plaques, makes the misleading "∼0.5 points on an 18-point CDR-SB scale" argument I refer to about antibody efficacy (a perfect drug could only have achieved ~1.5 points efficacy). Doesn't engage with the open-label extension evidence of widening benefits after the initial 18 months of treatment.
4. For treatment before cognitive symptoms which I argue for, relies on the clinical failure of amyloid antibodies such as solanezumab which also failed to clear plaques. (The ones which succeed at clearing plaques haven't finished the relevant clinical trials.)
5. Frames the (I agree very bad) fraud as pertaining to "core Alzheimer’s findings that had guided decades of research and many hundreds of millions in funding", which isn't really true, as I discuss.
6. Misunderstands the lower soluble Aβ42 in most Alzheimer patients' cerebrospinal fluid, saying it's a new finding that challenges the amyloid hypothesis. But we know that it's due to impaired clearance (the missing Aβ shows up in amyloid plaques on PET scans), and in fact for this reason it's been considered a standard biomarker of amyloid pathology for decades.
I shared this with an expert in the field as I have not kept up with the literature since retirement in 2022. Her comment was: "He actually did a great job distilling a complex picture and history into a very current and comprehensive view, and provided a nice bibliography to go along with it." It was interesting to see the exciting preliminary results from the Trontinemab trial. You have positively changed my bias point about future Alzheimer's treatments and provided a very readable example of the way science slowly, but positively progresses. Thank you for this excellent writeup that has helped to bring your readers up-to-date.
It’s impressive the author has made a specific and time-stamped prediction about the progress of clinical trials in this field. So often people “predict” without standing up for it.
Ultimately, biological plausibility is great but it is no substitute for clinical outcome trial evidence. So for now, it remains a hypothesis awaiting confirmation.
This term is used by different people in different ways, but on this blog, the usual concern about “AI safety” is “how can we make sure that we don’t end up with an AI that is so unsafe that it leads to human extinction?” This term has been partly overloaded for lesser things, like “make sure AI does not use naughty words”, so some people have looked for terms that are less likely to be “stolen”. The current leading candidate is notkilleveryoneism.
You’re welcome! There were some posts on the old blog that talked a bit about this; I feel like almost a quarter of those tagged “ai” at least mentioned it: https://slatestarcodex.com/tag/ai/
The new blog has some too but they’re harder to find because substack doesn’t have tagging. (To be fair I’m not sure Scott was perfectly consistent in tagging on Wordpress but still.)
I've suspected for years that I'm having mild cognitive decline, but I'm under the impression that there is no affordable test that would test positive years before I fail any cognitive test. And since I'm "too young" to get Alzheimer's, I expect no doctor would order an expensive or invasive test. Given that companies are still pursuing Aβ treatments, is there a good, safe, affordable test in the pipeline? If someone is very interested in getting a test, is there some plausible way to get one?
P.S. twice now I've experienced sudden partial vision loss, which I interpreted as a potential stroke (both eyes affected equally; parts of visual field replaced by "snow"; fovea unaffected; the "snow area" slowly changed shape; problem disappeared within an hour). Before that, I woke up one day and noticed that some sharp corners on fonts and shapes looked rounded, although I looked closely to confirm that they were, in fact, sharp corners. I figure if my visual system is suffering minor damage, the rest of my brain probably is too. These phenomena seem like mini-strokes rather than Alzheimer's, but I don't know anything I can do for diagnosis or treatment of mini-strokes either. My doctor seemed unconcerned, so nothing was done.
A p-tau217 blood test (Quest Diagnostics) is $1264.00 at Ultalabtest.com. My results were back in two days. There were multiple phlebotomy sites to choose from.
Thanks. From what I can tell by a quick search, they make no representation about how early in the progression of AD the test should turn positive or "indeterminate". Not surprising for a non-FDA-approved test, but for that price I'd need a test that reliably detects AD very early.
David, for your sake please contact a neurologist and get yourself screened to rule out multiple sclerosis. The optic nerve is commonly one of the first to be affected in a way that the patient notices.
Among the many early signs of MS are the vision problems "optic neuritis, double vision (diplopia) involuntary eye movements (nystagmus)". I don't have these symptoms.
That was a truly amazing post. Well written, well reasoned, and well sourced. I followed the link in footnote [124] and was amazed again, this time with horror. I hadn't realized it was that bad.
Thank you. The fire analogy really clarifies the stakes and the timeline. The scary part is when you map the 10-20 year preclinical window onto demographics. It means the fuse for the baby boomer retirement crisis was lit around 2005, and we're now entering the critical decade to act before the house really starts burning down. A "Warp Speed for Alzheimer's" similar to the deployment of the COVID vaccine feels like the obvious response, because the alternative is letting it consume ~$1 trillion/year (per the Alzheimer's Association's projections), tying a lot of work force and loosing endless DALYs. I wonder which country will be the first to employ a concerted effort. I feel like my own country (Germany) might be better equipped than the U.S. currently. Feels like the first country to treat this as a national industrial strategy rather than a healthcare cost is going to win big.
Missing reference to Bredesen, FINGER, MIND diet. Amyloid hypothesis is not defensible. There is resistance to any alternative IMO because establishment researchers knew the faked studies and doubled down for ego defense. Establishment researchers who knew and cared about science, left the establishment and have been going at prevention for years and are now ahead by years. Acceptance of their peer reviewed research by the establishment amyloid researchers would be a blow to egos.
What I find particularly striking about this whole discussion is how much it reveals about the huge importance of social dynamics in the sciences.
It took you, a very smart and highly educated person, quite a long time to dig through these papers and even then you are highly reliant on the arguments and opinions about this area willing to be circulated in public.
I'm guessing you do have a pretty good picture of the state of the research but it wouldn't take much to shift the plaque hypothesis from disfavored but talked about to too embarrassing to mention in academic circles (eg the way geologists refused to consider a huge flood as the cause of the badlands as it felt too much like discredited biblical based geology) and you would have been unable to gather much supporting evidence.
And beyond this my sense is that a huge amount of the knowledge in a field is never exposed in published articles but merely conveyed by a sense amoung those who do the experiments of what does and doesn't seem to work. So we really are, probably unavoidably, extremely reliant on the social dynamics of experts in the field.
And no doubt this is why even sophisticated investors like drug companies and VCs find it unprofitable to fund many outsider ideas. I mean funding a view that's seen as less likely but still serious is one thing but the second you fall off that cliff it becomes almost impossible for anyone who isn't directly involved in the field to tell the difference between something that's a 1 in 100 chance of being true and 1 in a million.
Great writeup. I wonder if anyone has ever tried injecting anti-amyloid antibodies directly into cerebrospinal fluid (intrathecal delivery) to bypass the blood brain barrier.
Also, regarding γ-secretase modulators, a lot of work went into these in the early 2000s and they ended up making things even worse due to side effects. γ-secretase is involved in processing a lot of proteins, notably including Notch, and it's quite difficult to block APP cleavage selectively.
I'm not aware of any intrathecal amyloid antibody delivery in humans.
Regarding γ-secretase, yeah, so those were γ-secretase *inhibitors* and they did have a lot of off-target effects, because of the enzyme's large number of substrates. However, despite a wide variety of mutations, the Alzheimer-causing ones strongly negatively correlate age-at-onset with the number of large amyloid species they lead to. [1] So it's being looked at again, but via more precise targeting of just the amyloidogenic pathways, hence the term "modulator".
[1] Fernández, Sara Gutiérrez, et al. "Spectrum of γ-Secretase dysfunction as a unifying predictor of ADAD age at onset across PSEN1, PSEN2 and APP causal genes." Molecular Neurodegeneration 20.1 (2025): 48.
>I wonder if anyone has ever tried injecting anti-amyloid antibodies directly into cerebrospinal fluid
That was my first thought also. If I'm faced with the probability of dying a horrible death due to dementia, you can go ahead and inject whatever the heck you want into my spine using whatever experimental methods you want. It can't be worse than the alternative, right?
Maybe it would if you had to do it 15 years before first sign of decline.
> injecting anti-amyloid antibodies directly into cerebrospinal fluid
I wonder if a better approach might not be nasal administration of amylophagous protists. I believe this has successfully been demonstrated as an effective route, with neti pots and "brain-eating amoebae."
Or just train "brain-eating amoebae" to target amyloids. "Sit." "No! Sit!" "Now sick-em!"
I considered this, but I think helminths would be better.
Because of their innate trainability? Or because their puppy-like demeanor would make the training process so much more fun?
Both good reasons, but my motivation was that you could get Yeerks.
Alzheimers! I'll take the Alzheimers, thank you.
::is not knowledgeable enough to know how much of a joke this is::
Depends who's asking. If the country code is +46, I'm deadly serious.
Those antibodies would still have to get from the subarachnoid space into the brain parenchyma.
These phase 1b tau trials are interesting. They were not designed to measure clinical efficacy but still it would have been nice to see some hint of it. Apparently we were not so lucky:
"Performance on functional, cognitive, psychiatric and neurologic clinical outcomes slightly declined as expected for participants with mild AD over the duration of the treatment and post-treatment periods." [116]
Oh, wait, "favorable trends on multiple exploratory endpoints of cognition and activities of daily living in AD (n=46)"
https://investors.biogen.com/news-releases/news-release-details/new-data-biogens-investigational-antisense-oligonucleotide-aso
Thank you for this! What do you think about that new study on lithium deficiency and Alzheimer’s? Seems like it relates to plaque formation
https://www.science.org/content/blog-post/lithium-deficiency-and-alzheimer-s-disease
I second your question and re-up my post from the last open thread
https://www.astralcodexten.com/p/open-thread-394/comment/144263199
I had this question as well, and am still interested in David's response, but I found this discussion of that paper by Derek Lowe pretty encouraging:
https://www.science.org/content/blog-post/lithium-deficiency-and-alzheimer-s-disease
Had a chance to spend some time with that paper today, but only a few hours, and I'm otherwise not super familiar with lithium/Alzheimer research (sadly, even with all the time I've spent, the field is so sprawling that there are plenty of topics I've only scratched the surface of) so take this with a (micro)dose of (lithium orotate) salt, but here's my understanding of their claimed results and my reaction:
- In both amyloid and amyloid+tau mouse models, many lithium salts (but not lithium orotate) co-accumulate with amyloid plaques, in a manner which appears competitive with other cortical locations, so that the presence of amyloid plaques seems to result in a reduction of lithium in cortex (but not a reduction in non-cortical lithium) relative to normal mice. [One caution so far: there's a LOT of crap found in amyloid plaques. So finding lithium there may or may not by itself be of significance.]
- Human Alzheimer patients show lithium concentration in amyloid plaques, and reduction in cortical lithium (but not elsewhere) relative to non-Alzheimer humans, similar to the mouse models. (It's not totally clear to me if the "Alzheimer" patients all actually have Alzheimer pathology as opposed merely to clinical symptoms. I'd like this clarified.)
- In mice, a lithium diet deficiency accelerates the emergence of Alzheimer-like pathology and cognitive deficiencies in mice already predisposed to them, but even a small amount of lithium orotate (but not lithium carbonate, which binds better to amyloid) is protective.
So there are two potential causal pathways they're identifying:
(1) Amyloid pathology (known to be in part due to amyloid overproduction due to the mouse models being used) -> less lithium available to cortex.
The purported mechanism here seems clear: the lithium is preferentially binding to the amyloid plaques rather than residing in other cortical tissue.
(2) Less lithium available to cortex -> more amyloid and other Alzheimer pathology.
In *this* case, I'm confused about what the purported mechanism would be. I can understand how more lithium with amyloid affinity could cause more amyloid pathology (by promoting further cross-aggregation), but what they're showing here is subtly different: it's that a *deficiency* of lithium that *doesn't* have amyloid affinity is causing more Alzheimer-like pathology. Perhaps it's an interaction with a third, unspecified molecule, pulling it away from amyloid and preventing the promotion of aggregation? Would be good to understand this better.
Theoretically, it seems like there could also be a feedback loop between (1) and (2), since they seem to run in opposite directions, but I'm not sure.
A few questions I'd like answered:
- What is the timeline of cortical and amyloid-associated lithium in humans with preclinical Alzheimer's disease (i.e. with amyloid pathology but not yet at the symptomatic stage)? They didn't look at this, but knowing the answer might help to rule out hypotheses about causality.
- What is the purported mechanism by which a cortical lithium deficiency induces Alzheimer-like pathology? This sort of question can be pretty hard to answer because it's common to identify candidate molecular pathways in vitro or in animal models that end up not being the important ones in the human disease.
- Do many human beings actually suffer from a (non-amyloid-binding) lithium deficiency in our diet? If so, then that increases the relevance of these findings, at least for some of us. If not, then they may have just uncovered a theoretical pathway to amyloid pathology and Alzheimer's disease which isn't relevant to most human beings.
I don't know if it will go anywhere, but it seems interesting and worth exploring more. There were some previous Alzheimer lithium clinical trials which mostly don't seem to have panned out, but they used different lithium salts which bind better to amyloid than lithium orotate, so it's possible that this would achieve different results. Apparently they are investigating the appropriate dose range for human beings, which currently isn't clear, before pursuing clinical trials.
Fascinating. Thank you so much for looking into this!!
In the section of the paper titled Lithium and Microglial Function, they describe transcriptional changes to lithium-deficient microglia, which are immune cells that usually help clear amyloid beta from the brain - the last line of that section summarizes their findings as "Li deficiency leads to a reactive pro-inflammatory state and impaired Aβ clearance".
So I think at least part of the causal pathway here is something like: 1) amyloid starts to build up and form plaques, which start to entangle lithium and reduce cortical lithium levels, which 2) impairs microglial function, reducing their ability to clear amyloid (and increases inflammation), leading to more plaques, and so on in a vicious feedback loop.
Thanks! This seems like a plausible mechanism.
Thanks for highlighting this.
Are there any of these drugs you would take yourself in (say) your 50s in an attempt to slow down the progression to Alzheimers? It seems like the endgame of this research is likely to be that approximately everyone 50+ takes an anti-amyloid drug along with their statin, in order to delay the onset of mental decline from tau long enough to let them die of something else. And if that prevented most cases of Alzheimers and didn't have terrible side-effects, that would be a really good outcome.
If I were amyloid-positive (as determined by plasma, cerebrospinal fluid, or PET scan) and pre-symptomatic, I'd try to enroll in the trontinemab prevention trial, which looks like the best bet right now. https://www.roche.com/media/releases/med-cor-2025-07-28
Trontinemab looks to have higher efficacy and much better safety than the on-market state-of-the-art.
I was going to ask this question too. I have one of the APOE mutations and I have a PET scan from a couple of years ago as a result of a cancer diagnosis experience.
Would it make sense I wonder to ask someone to look at my PET scan through this lens and then if there is evidence, would that qualify me for a prevention trial?
Your PET scan wouldn't have been with an amyloid tracer, so it won't be helpful here. If your only genetic risk factor is ApoE4, you're very unlikely to be in the preclinical phase until at least your late 40's or early 50's. If you are of that age, you could probably get a blood, CSF, or plasma test for amyloid pathology and then if that's positive, attempt enrollment in one of the prevention trials I mentioned (which will also confirm the pathology before enrollment).
Thank you for clarifying that. I'm 60 actually. My brother has Lewy Body dementia, but I gather that variant is not genetic in terms of raising my risk.
Sometimes the term "genetic" can be used confusingly to just mean autosomal dominant (one copy guarantees the disease), even if there are risk alleles. ApoE4 is also a risk factor for LBD, but I know much less about it.
Hi Radar.
I work in the field and just want to second David's advice here. If you're 60 and you have an ApoE4 gene there is a good chance you have pre-clinical AD, and getting an understanding of your amyloid levels now (and really every year from now to monitor the change if you're not ready to take potentially risky treatments) will be beneficial.
If I could afford it I would have a PET scan. If not, probably CSF although a spinal tap is hardly much fun. The plasma tests may not be accurate enough for monitoring yet.
Omroth, I really appreciate your taking the time to add your thoughts from your expertise.
Because of this post and your and David's comments, I've reached out for a referral to a neurologist to talk about all this, so thank you very much for the nudge. A yearly PET scan given my radiation load (BC cancer treatment) and risk of cancer recurrence sounds risky, but I don't know the details on that.
Another factor to my risk profile is that one of the chemo agents I was given also has some maybe correlation to higher AD risk, though the research looks equivocal. I don't know if you know anything about that?
The other thing I worry about is that the aromatase inhibitor I have to take makes sleep less good. I take things to make sleep good, but the need for amyloid clearance at night feels important.
Up until recently I was someone who barely took Tylenol and had no health issues, and I avoided basically all medical care. It's been wild to enter into this world of extensive medical intervention and now to be kind of enthusiastic about it (not about getting intervened on, but about what's possible at the newer edges of what's available).
If you’re worried about it, it would be good advice to get enough sleep, and regularly. Stage 3 sleep seems to be the time when vascular fluid “washes out” the brain, potentially removing proteins before they become aggregations. Also it’s good to get sleep in general, so you won’t be wasting your time even if it doesn’t end up mattering for Alzheimer’s
> High intelligence seems to be a protective factor (hopefully this is reassuring to a lot of people here)
I don't think this is actually true. Rather, I think high intelligence is characterized as "protective" because, after suffering a lot of loss-of-function, formerly high-intelligence people still look normal to low-intelligence people.
But often, high intelligence people engage in more intellectual activity, much like physically gifted people are more frequently athletes. This continued intellectual activity late in life may be ,even if not "technically protective," still effectively protective by continuing to build new function, even as old function degrades.
Is there evidence that there is causation running from these to lower Alzheimer's risk? Assuming there's actually a correlation (Idk if there is), for several of these it sounds very plausible that the direction of the causation is that early Alzheimer's causes old people to stop maintaining strong social connections etc. unless there have been studies to disambiguate this from the direction of causation you claim.
I didn't heap criticism, I asked a question.
I'm not sure whether people are just not allowing enough time for stage 3 sleep, or if people at risk for Alzheimer's don't have a good capacity for stage 3 sleep.
Thanks! That was a great overview and clarified a lot of aspects for me. In particular, I didn't know that the tau folding is so specific to its cause.
Thanks, great writeup.
My background is that I'm a working neuroscientist in a field far from Alzheimer's research. From talking with colleagues, I'd often gotten the impression that amyloid was a dead end. So it was helpful for me to read the section titled "Challenges translating from mouse models." With that context, it's clear to me that a lot of the pessimism among neuroscientists stems from stories of those early, poorly-conceived mouse experiments.
Neuroscience research is often plagued with an unhealthy optimism that if you squint hard enough at an experiment in a mouse, it will seamlessly provide a good model for what's happening in humans. There's a large contingent of scientists ready to jump on subfields that produce large quantities of thoughtless work that suck up attention and available funds. This was clearly appropriate for some Alzheimer's research, but not for the field writ large.
This was cool, but I suppose it won’t be testable for a long time. A shame, because Alzheimer’s is a really bad disease
Yes. IIUC, the side effects of the existing (approved) treatments for Alzheimer's are sufficiently bad that you're just as well off not taking them. Which is a pretty thorough condemnation.
I, too, am not an Alzheimer's researcher so I defer to my betters--specifically, Derek Lowe's extensive reporting on the anti-amyloid therapies in his blog at Science. Lowe did work in the field and has been following it throughout this career: https://www.science.org/content/blog-post/does-it-work-does-it-do-harm-and-more-basic-questions.
A quick summary: The much-touted “27–33% slower decline” with lecanemab and donanemab masks the reality that these effects are clinically trivial, well below the minimum change patients or caregivers can perceive. Even donepezil, widely regarded as only modestly effective, outperforms these numbers. Trials last just 18 months, leaving it unknown whether these gains (imperceivable to caregivers) persist.
Meanwhile, risks are substantial. Amyloid-targeting antibodies cause brain swelling, microhemorrhages, and deaths, with the greatest hazard in APOE4 carriers and women. The APOE4 risk association requires some unpacking to be sure! See here: https://www.eneuro.org/content/11/7/ENEURO.0319-23.2024
Proposals to treat everyone with amyloid or tau under the expanded ATN criteria would expose millions of cognitively normal older adults (most of whom will never develop dementia) to these dangers. See here: https://www.alzforum.org/news/conference-coverage/revised-again-alzheimers-diagnostic-criteria-get-another-makeover
But proponents of treating earlier in the disease course have already tested this hypothesis to disappointing results: The ambitious A4 trial tested this idea by treating older adults who had elevated amyloid but no cognitive symptoms. For 4.5 years, participants received solanezumab, an antibody targeting amyloid. Participants treated with solanezumab saw no cognitive benefit whatsoever—in fact, they performed slightly worse than those who received placebo. See here: https://www.nejm.org/doi/full/10.1056/NEJMoa2305032
Through a different mechanism in the late 1990s, this hypothesis was also tested with the AN-1792 vaccine that cleared plaques but did not halt progression to severe dementia. The trial was halted early due to severe brain inflammation and post-mortem examinations revealed a profound disconnect: despite dramatically reduced amyloid plaques, these patients had continued to develop severe dementia anyway. See here: https://academic.oup.com/brain/article/142/7/2113/5510133?login=false
The field’s response to repeated underperformance has been to shift diagnostic definitions and adopt weaker outcome measures (“time spent in mild stage”) that exaggerate benefit while obscuring minimal impact on real-world function. See here: https://www.nature.com/articles/d41586-024-00756-8
Many of these points are addressed in the article, but to respond directly:
Unlike donepezil, which only mitigates symptoms and doesn't affect the underlying biology (if you go off it, you end up where you would have been had you never been on it), the benefit from the partially-effective amyloid antibodies seems to get larger over time (open-label extensions point to widening benefits through four years), and persists vs. placebo even if you go off them. Despite this, they're not where we'd like them to be on either safety or efficacy, and I offer my belief as to why in the article: difficulty crossing the blood-brain barrier and treatment too late to prevent tau pathology. Thankfully both of these matters are being addressed.
I agree the side-effects can be serious, and are especially common for ApoE4 carriers. New BBB-crossing approaches like trontinemab seem to mitigate that risk greatly for everyone. A new donanemab titration schedule also appears to bring the ApoE4 carrier risk much closer to that of non-carriers: https://www.alzforum.org/news/conference-coverage/donanemab-small-tweak-titration-big-gain-safety
The solanezemab trial failed but this is one of the antibodies that didn't clear out plaques. In the study you link: "Amyloid levels on brain PET increased by a mean of 11.6 centiloids in the solanezumab group and 19.3 centiloids in the placebo group." (So a small between-group difference, but still a net increase. Whereas lecanemab, donanemab, and especially trontinemab cause deep plaque clearance.) The preclinical prevention hypothesis will be better tested by the clinical trials I mention in the post.
I had to read up on the AN1792 vaccine because I wasn't familiar with it. The main things I'd point out from that study are:
- The trial was interrupted because of meningoencephalitis in 6% of patients. [1]
- 5 of the 22 they examined post-mortem appeared to never have had amyloid plaques in the first place (sadly, it was uncommon back then to confirm Alzheimer's disease with biomarkers before clinical trial enrollment, which greatly reduced statistical power on many trials).
- They report that 7 of the remaining 17 only had "very limited" or "no" plaque removal.
- That leaves only 10 of 22 which had what they call "intermediate" or "very extensive" plaque removal.
- Treatment began after symptom onset, which is expected to be at best only somewhat effective under the ATN model.
- Despite this, there were hints of efficacy: p = 0.02 on their neuropsychological test battery for antibody responders. [1]
I wouldn't pin my belief on the amyloid hypothesis on this trial, but nor does it seem like clear evidence against it.
[1] https://www.neurology.org/doi/10.1212/01.WNL.0000159740.16984.3C
The anti-amyloid drugs we have now clear plaques more than enough to test the idea that removing amyloid will change the course of Alzheimer’s. That idea has been tested for decades, and it hasn’t delivered.
When you strip away the open-label extensions, which can’t prove durability because there’s no control arm and plenty of selective drop-out, the blinded trial data are all we can anchor to. The gains are statistically significant but far smaller than the minimal clinically important difference threshold, the point where patients or caregivers can actually feel a change. Even donepezil still compares at least as well in practical terms. Until a blinded trial shows a sustained, perceptible benefit, claims of “widening” or “persisting” effects are unwarranted.
Trontinemab may eventually offer a better safety profile, but right now that’s based on cross-trial comparisons and early-phase data. There’s already been at least one hemorrhagic death. We need fully powered, placebo-controlled Phase 3 trials before declaring the risk “greatly mitigated.”
And it’s worth remembering: even when AN-1792 produced substantial plaque clearance in half the treated group, those patients still progressed to severe dementia. That’s the core problem: namely, clearing plaques has not, so far, translated into meaningful change in the disease trajectory.
Instead of confronting that, the field keeps moving the goalposts. The latest proposed criteria would drop neurodegeneration entirely from the definition of Alzheimer’s and rely only on amyloid and tau biomarkers whose causal role remains unproven. Under that framework, many cognitively normal, amyloid-positive older adults would be labeled as having Alzheimer’s, while a neuroprotective drug that preserved function without touching amyloid or tau could be branded a failure. Critics have called for definitions rooted in clinical assessment and direct evidence of neurodegeneration, with biomarkers treated as unproven surrogates until they’re linked to patient-felt benefit.
Isn't it worth being a bit skeptical as to why these new definitions are being pushed? I am not anti-pharma by any means but there is a awful lot of money at stake here and these definitions seem bonkers given that ~1/3 of clinically normal older adults show substantial amyloid plaque deposition yet never develop Alzheimer’s disease during life. Keeping neurodegeneration as part of the criteria seems essential in distinguishing between normal aging and disease.)See a review by some alarmed neurologists here: https://pubmed.ncbi.nlm.nih.gov/38104639/
Until a treatment can meet that bar, durable, meaningful improvement at acceptable risk, changing definitions or endpoints won’t change what matters to patients and families.
FWIW, I mostly don't think you're addressing the points in the article or in my first reply. But I'll reply once more:
- First of all, my position is not "the existing amyloid antibodies are great!" It's "the amyloid hypothesis looks right, but there's still a lot of work to do".
- Second, I want to reiterate that treatment before symptom onset seems crucial under the ATN model, which has a lot of evidence behind it (discussed in the article). So the eventual progression to severe dementia after amyloid clearance is consistent with the ATN model.
- Third, I went into this investigation skeptical of the amyloid hypothesis, as you propose one should be. I was very prepared to find it lacking. I came away persuaded.
- Fourth, while OLEs do suffer from lack of placebo, the selective drop-out problem also applies to the ADNI and BioFinder controls they compared against. There's not an obvious reason to expect its effect to be much greater for lecanemab/donanemab than those groups. That said, of course we'd like to see longer comparison just against placebo in an RCT, and I certainly would concede that evidence from two OLEs is not by itself conclusive.
- Finally, in the absence of such long-term placebo-controlled trials, your position has less evidence behind it than mine. You're claiming we know that the drugs provide a clinically meaningless benefit because you're *assuming* that the effect does *not* widen, even though we 1) know the effect is disease-modifying, not merely symptomatic (since when you go off the drugs the effect persists, unlike with donepezil), and 2) have OLE data which at least offers some evidence, even if it isn't conclusive. Indeed, even if the drugs were 100% effective at slowing progression in those 18 months, the benefits would still be very dubious by your logic, since you're assuming (without empirical backing) that they won't continue.
Thanks for the conversation, I think we've made our arguments and let's revisit in 10 years:)
... just for those slower ones in the crowd, ... um, what do you think about the ATN model?
it seems to me that arguing about how some idiots want to change the definition is not strictly relevant, no?
or, your last comment implies that we simply don't have enough high quality data to really test the ATN model?
because to disprove it we would need to get increasing-A and no-T (and no-N and no-AD) people, put them on A-clearing stuff, and wait, *and* if they still get T *and* N (and AD) then it was not the A?
or you are saying that the fact that millions of people have a lot of A and no AD (and so likely no N) means that the whole model is poppycock?
thanks!
Thanks for the question.
First, on legitimate criticism of amyloid primacy:
The skepticism isn't coming only from "no-nothing outsiders" but from prominent neuroscientists and drug discovery experts. Derek Lowe, a respected medicinal chemist with decades in pharma, has chronicled the field's endless moving of goalposts. As he notes, it's always: "We haven't dosed early enough, in the right patients, in the right way, targeting the right sort of amyloid." This refrain has continued for 30+ years while clinical benefits remain imperceptible to patients and caregivers. https://www.science.org/content/blog-post/had-enough-eh
On testing the ATN model:
The model makes two core claims about amyloid:
• Necessary: No Alzheimer's without amyloid
• Sufficient: Amyloid alone drives downstream pathology
For necessity, the definition now tautologically requires amyloid, but Primary Age-Related Tauopathy (PART) presents a challenge as these patients have tau tangles and can develop dementia without significant amyloid. The field debates whether this "counts" as AD, which reveals the circular reasoning problem.
For sufficiency, approximately 1/3 of adults over 70 have substantial amyloid burden yet remain cognitively normal. I’m not quite sure on the author’s position here but seems to argue these people would "eventually" get AD if they lived long enough: an unfalsifiable claim since many die cognitively intact at advanced ages. This breaks the sufficiency assumption.
Why definitional changes matter:
The proposed criteria would diagnose AD based solely on biomarkers (amyloid/tau) while dropping neurodegeneration and cognitive assessment. This has massive implications:
1. Medicalizes aging: One-third of cognitively normal elderly would be labeled as having "Alzheimer's"
2. Changes trial endpoints: Success becomes reducing biomarkers, not improving function
3. Creates treatment imperatives: Millions of healthy people become candidates for as of now risky, expensive drugs
As concerned neurologists note in the link I gave above: "Everyone agrees that AD is a form of cognitive decline caused by neurodegeneration, but neither cognitive decline nor neurodegeneration are tested for or given diagnostic value in the revised ATN system."
The public health nightmare in my view:
If we accept that treatment must begin at first amyloid detection (before tau), we're proposing decades of prophylactic treatment with:
• Unknown long-term consequences (antibodies are notoriously promiscuous in binding; what are the off target effects over years?; antibody therapeutics can be especially problematic in people with autoimmune risk, so what additional harms could we be causing?)
• Documented serious adverse events (brain swelling, hemorrhages)
• Enormous costs ($26,000-56,000 annually per person)
• No way to distinguish who will progress vs. remain healthy
Alternative frameworks:
What if amyloid accumulation simply reflects impaired protein clearance with genetic susceptibility overwhelming the equilibrium decades earlier? The glymphatic system's role in brain clearance has only recently been appreciated. Sleep disruption, vascular changes, and inflammation all converge on clearance pathways. This suggests targeting clearance mechanisms rather than treating everyone with detectable amyloid for decades. This is just one alternative therapeutic target that can assemble the evidence into narrative cohesion, there are many more to be plumbed.
The ATN model may capture one pathway to neurodegeneration in susceptible individuals but treating it as universally deterministic ignores the millions who age successfully despite amyloid. The push to redefine AD around biomarkers rather than clinical syndrome serves commercial interests more than patients. All we can do now is return in another decade to see whether the next iteration of "earlier, better, more" finally delivers or whether we're still hearing the same refrains.
As someone who both worked on the Falcon 9 and on Alzheimer's disease, how often do you make "it's not rocket science"/"it's not brain surgery" jokes?
I'm partial to "it's not rocket surgery".
I tend to say that rocket science is high school physics. It's rocket *engineering* that's a pain.
It's neither. It's rocket *plumbing* that's hard. If the plumbing works, the science and engineering are generally A: straightforward and B: fun even when they get tricky. But rocketry is pretty much a worst-case environment for plumbing nightmares.
Most of what you say here sounds reasonable.
I still think it's probably a mistake for drug research to target amyloid. Amyloid is produced because it has many benefits. So it seems somewhat likely that the high amyloid levels are a mostly desirable reaction to some other problems, and more attention should be devoted to the problems that are upstream of amyloid.
If this were the case, we'd expect ApoE3 homozygotes — and others who are not at much higher risk of Alzheimer's due to increased amyloid production or impaired clearance — to be dropping dead of infection left and right, or at least frequently enough to outweigh the much reduced risk of Alzheimer's. But we don't see that.
As a former (systems, not molecular) neuroscientist, I don't find this terribly convincing? Claims like "gene X does only Y" have a very very poor track record, and brains of a certain genotype may have perfectly good reasons for what seem to us like inexplicably excessive amyloid levels. This is also consistent with reports of negative side-effects from existing drugs.
Having said that, I think we must keep trying on amyloids and dealing with the consequences as they come.
It’s not so much “gene X does only Y”, but more “Y is the only plausible mechanism that all the risk genes have in common, and for some of them the only plausible mechanism that has been proposed".
That's a very reasonable claim, but I don't think it is a sufficient answer to Peter's objection.
I think you can answer Peter by saying "well, Alzheimer's is very bad, so any problems we cause by clearing amyloids are likely to be less bad than Alzheimer's, and anyway over time we can minimize them with calibration/optimization/compensation".
I don't see how you can get a clear prediction about Apoe3 people dying from infection without making some assumptions that I consider to be questionable.
For example, some of the reduced amyloid with Apoe3 could be the result of Apoe3 handling pathogens better, so that infections less frequently reach the stage where amyloid is called for. If that effect is important, I'd expect Apoe3 people to die from infections less.
Apoe variants have important effects on infectious diseases. I suspect that amyloid-related differences only cause a small fraction of these effects.
I also note that there are diseases for which Apoe3 is clearly worse. So your observation that we don't see Apoe3 people dying more is fairly dependent on where you look.
The more I look, the more complex this topic seems.
OK, so I understand your view to be that ApoE4 carriers are worse at fighting off infection, and this means that infections in ApoE4 carriers are more likely to reach the point where the brain resorts to amyloid as a defense mechanism, and that while this defense mechanism then leads to Alzheimer’s in many cases (~80% lifetime risk for homozygotes), it’s still worth it compared to the alternative, which is a very bad infection. Is that right?
If so, a few points about this:
1. Alzheimer’s is so bad that the infections in question would have to be extremely bad or extremely common to justify that tradeoff. While this is possible, I’m not aware of any strong reason to believe this would probably be the case.
2. This seems to predict that those treated with aducanumab, lecanemab, donanemab, and trontinemab, all of which clear out amyloid plaques pretty well, would get many more and much worse infections than those treated with placebo. But there is no hint of this in any of the safety data: the main frequent and serious adverse events are brain bleeds and swelling related to the unintended congregation in blood vessels; and trontinemab, which mostly avoids this, has much less bleeding and swelling while achieving much better amyloid clearance.
3. You’re not eliminating all amyloid, only the fibrillar plaques, and only in people who are very likely to progress to dementia without treatment because they already have an amyloid burden which would cause that. This means the antimicrobial properties of soluble amyloid species, if they are very important, can still mostly function (just as they managed to function for the first ~50 years of the person’s life without severe amyloid plaque burden). And once you clear out the plaques, treatment might not be necessary again for years (amyloid plaque burden grows pretty slowly, including after clearance with antibodies).
It's quite plausible that overproduction of ANY of the body's proteins/hormone/etc. can be hazardous. It's what you should expect. And you should also expect every system to be interconnected with lots of other systems in ways that aren't obvious.
I truly expect that if you could clear out the (excess?) amyloid plaques without doing other damage it would be beneficial. Also that if you could eliminate the tau misfolding it would be beneficial. And it's quite plausible that these are connected. I also have a suspicion that the tau misfolding is more directly causal of Alzheimer's. If the amyloid causes tau misfolding, then removing the plaques should not be expected to solve the problem after it has initialed tau misfolding, because I believe that they tau misfolding, once initiated, is a prion disease. But it might well slow it down by halting the creation of new roots of misfolding trees.
Aducanumab, Lecanemab, Donanemab, oh my. Since they all end in ....mab I'm assuming there is some method to drug naming, does anyone have a pointer to the decoder ring? I had always just assume pharmaceutical companies had terrible marketing departments.
-mab means it is an antibody therapy, -ib is a small molecule inhibitor (typically kinase inhibitors). -vir are antivirals. More naming conventions: https://en.wikipedia.org/wiki/Drug_nomenclature
Thank you very much. This helps...some.
mab is for monoclonal anti-body.
you get used to it. except with cell and gene therapies — a lot of their official generic names are so hideous no one even attempts it and short nicknames are universally used instead. i once came across a conlang someone had constructed wholly out of drug nomenclature, and spent an afternoon both 1. elated by it and 2. writhing in frustration, because none of the lingustic design choices actually corresponded to or were meaningfully rooted in the drug nomenclature? i just really wanted there to be, idk, assignments of such and such a drug class/suffix to such and such a bit of grammar or syntax that *made sense* lolsob
Thanks, this is fascinating - I always assumed drug companies were just going the Chinese junk brand route of keyboard mashing.
Ha ha, we need a name for our new drug, fire up the strong password generator and grab the first one that comes up.
These are scientific names, like "acetaminophen", not commercial names like "tylenol". Commercial names are the ones that sound like they come out of marketing departments... most likely because they do.
Funny you use that Tylenol as an example, got me interested so I had to look it up on wiki: "Like the words paracetamol and acetaminophen, the brand name Tylenol is derived from a chemical name for the compound, N-acetyl-para-aminophenol" I can't figure out how to add bold font here but wiki has bold for the tyl at the end of acetyl and enol from the last four letters of aminophenol.
MAB = Monoclonal AntiBody
Reading those always makes me wonder if they are actually reversed names. Itd be funny if their eventual brand name is indeed its clinical name reversed.
I guess my question is: if this is a disease caused by amyloid overproduction, and we know which genes cause amyloid overproduction, can't we just do a CRISPR thing and, like... turn those genes off? Turn them down a bit?
There is a phase 1 trial for a drug to turn down amyloid production: https://capella.alnylam.com/wp-content/uploads/2025/07/AAIC-Phase-1_OP_SUBMITTED28Jul.pdf
Note that *most* instances of the disease are caused by impaired clearance, not overproduction. However, it is still possible that turning down production can help restore the balance.
Can some of the actual neuroscientists and medical practitioners chime in about having "smart guy from SpaceX" enter into the medical field of deciding what constitutes promising research after doing six months of reading?
This was a great piece and it seems convincing, but I don't know the field.
The last time a bunch of smart guys from SpaceX got into medical research.... It destroyed the NIH. I know some people forgot this ancient history because it was over a month ago, but a little bit of "I'm sorry that some of my former coworkers at the behest of my former boss cancelled funding for vast amounts of promising therapies in all medical fields, but here is my piece."
The idea that David Schneider-Joseph has worked at SpaceX in the past seems like interesting background but not relevant to his post. If he hadn't worked at SpaceX but had written the same words would that make his post more compelling?
Part of what we look for in anyone is "do we trust their background and who they are". We licence physicians and lawyers as a way to tell non specialists that the licensed individuals have a certain degree of training.
OK, so him disclosing not being a physician or neuroscientist is good.
If his previous job was being a plumber, that would actually be better.
But his previous job was SpaceX. Elon Musk's recent conduct at DOGE was to empower some very bright people (in his estimation) to take a blowtorch to medical grants that literally tens of thousands of researchers and millions of patients were depending on.
Why? Because Elon Musk is a smart guy. He and his underlings could decide what was good medical research... Trust them, they are smart and many of them worked on rockets!
So reading this, I am seeing "former SpaceX engineer says he knows better than medical professionals." That is a super negative prior for me personally based on what other people with similar biographies just did to American medical research.
I'm sorry that some of my former coworkers at the behest of my former boss cancelled funding for vast amounts of promising therapies in all medical fields, but here is my piece.
(In all seriousness, I'm very upset with what DOGE did.)
I write this not to tell Alzheimer researchers what they should be doing, but to convey what I learned about the field to other people, like myself, who are not in it.
It might be that you are not a lifelong medical researcher that contributes to the high readability of this piece. You've obviously learned an enormous amount about this issue in six months, and maybe it's your ability to think like someone not embedded deeply in the research that helps you translate what you've learned into such a readable piece for the rest of us.
I do appreciate you saying it.
I really like the piece and I find it pretty convincing.
I dunno! If somebody close to DJT reads this you may end up in charge of all future Alzheimers research funds. Be prepared!!
> I am seeing "former SpaceX engineer says he knows better than medical professionals."
I know the author has already replied, but I want to point out I think there's a serious basic flaw with this observation.
The Amyloid hypothesis is still the orthodox position in the field. It's more contrarians and outsiders who have claimed that it's false. So my read is that this post is more "here's why the experts ARE right.
But actually, more basically: When there is disagreements amongst experts, you can't agree with *both* sides. So that means EVERYONE who is convinced by one side or the other will find themselves believing they know better than (some) experts; it's unavoidable, unless you're willing to hold off on making conclusions until and unless there is consensus in the field.
What he's actually saying is more along the lines of "hey, it looks like the experts were on the right track all along, despite the superficially believable arguments given by contrarians. This is a hard problem, we shouldn't abandon what they're doing just because it hasn't yet had positive results yet."
Man I wish the people at Doge were this thoughtful.
Well, Scott posted this, so presumably that's at least a once-over from him.
It's true that skepticism is warranted when people Do Their Own Research™ outside of their area of expertise, but for really high quality people it's not impossible for it to turn out well. I mean, what, do you want make it taboo for non-doctors to engage in any sort of public discussion of medical research? All that's necessary is for the person to readily disclose that they don't have formal training, which he did, and for the audience to keep that in mind while reading.
>a little bit of "I'm sorry that some of my former coworkers at the behest of my former boss cancelled funding for vast amounts of promising therapies in all medical fields, but here is my piece."
That level of guilt by association is not remotely healthy. He has no obligation whatsoever to preface an article completely unrelated to Elon Musk with something like that. You are trying to bully someone into going out of their way to help you spread your political opinions. Cut that out - in general, but especially in ACX comments.
Well said!
Amen to that.
Former neuroscientist: this seems fine? Specialists will have their careers and egos and grants tired up with very narrow sub-problems in Alzheimer's research---in this case, for instance, a particular dead-end mouse model---so a disinterested outsider may have an easier time seeing the big picture. Just obviously don't take it as gospel.
In terms of the substance of this post, the main red flag I noticed was this statement: "The clearest relationship is various genes which massively increase amyloid production (while doing nothing else)". In neuroscience, claims of this sort are almost always wrong and must be viewed with intense suspicion.
I agree it may be close enough to being true for the problem at hand, and that specific claim wasn't especially load-bearing for the rest of the post anyway, which does a nice job laying out the convergent evidence. I probably should have called it "the only yellow flag" rather than "the main red flag" because it really was a very good post.
Responding late to your edit, but my point wasn't that APP and presenilin or any other biomolecule have other specific known functions; it's that in biology, you will never go broke betting that there is another function.
The practical implications here are... not much really. The amyloid hypothesis laid out here sure sounds like the best bet available. I'd just downgrade the probability and/or expected degree of success.
Good to hear it passed the smell test. As a non-scientist who has worked in drug and device development, it struck me as lacking epistemic humility. The track record of non-specialists wading into biology is not great, often because there’s a tendency to a) over-index on a single hypothesis and b) over-index on hypotheses that can be understood by a non-specialist. It would be much more credible if it covered all major hypotheses being explored and the evidence for or against them.
I am not a neuroscientist but I am a biomedical researcher who has followed Alzheimer's research for years -I think David did a good job summarizing the research consensus in a readable fashion, and I have no problem with this in general. Often researchers deep in the weeds have trouble explaining their field in a way that is easy for outsiders to understand, and they often are too busy with research to do even if they have the ability. In my own field of virology, I think Peter Miller did a better job than any virologist so far (including myself) of explaining the evidence for and against the lab leak hypothesis, and I've seen many colleagues with the same sentiments.
The real tricky stuff comes when an outsider with no relevant expertise tries to *disprove* a widely accepted hypothesis, instead of summarize a scientific consensus...still not out of bounds, but the odds of success are...much lower.
> The real tricky stuff comes when an outsider with no relevant expertise tries to *disprove* a widely accepted hypothesis, instead of summarize a scientific consensus...still not out of bounds, but the odds of success are...much lower.
Nice heuristic; I'm planning to still this summarization.
>Can some of the actual neuroscientists and medical practitioners chime in about having "smart guy from SpaceX" enter into the medical field of deciding what constitutes promising research after doing six months of reading?
I agree with this general sentiment. SpaceX/DOGE controversies aside, six months reading literature in your spare time is not nearly enough time to really understand a large, sprawling, comprehensive field, no matter how bright you are. You wouldn't ask for an authoritative review of a field from a first-year PhD student, and they aren't even working another job full-time!
I don't want to detract from this piece, nor David's abilities -- it's very well-written and rigorously researched. And I don't want this to turn into "Trust The Experts (TM)"; I think generally it's fine / useful for smart and capable people to write deep dives on some topic they get interested in, but this is exactly the kind of topic where a true credentialed expert in the field (not Scott, not David, not me) is the only person who could persuasively write "the defense of the establishment view on amyloid hypothesis." I have just seen too many examples in my own field of biomedical research where smart outsiders get one-shotted by first-year PhD-level pitfalls to fully trust something not written by a credentialed expert.
That David should apologize for the actions of some of his ex-colleagues working in totally unrelated roles seems wild to me. He made a wonderful effort post here that's very well footnoted and seems like a gift to all of us.
"The last time a bunch of smart guys from SpaceX got into medical research.... It destroyed the NIH"
David Schneider-Joseph is not those guys, and I'm disappointed to see that sort of drive-by guilt-by-association smear in this forum.
Did he not site enough papers for you? Are you looking to be certain he is correct without bothering to read any of the dozens of references? Why not just make a Bayesian update to your priors and move on....
You mention "extremely rare protective genes" that can protect some people even after the amyloids take hold. Has there been any research into trying to copy those genes' effects for a later stage treatment?
Lexeo is trying to deliver protective ApoE variants via an AAV vector, and in some drugs (LX1020) also suppressing expression of the pre-existing harmful variants: https://www.lexeotx.com/programs/cns-programs/
I think the evidence is pointing to the harm (at the tau stage) mostly coming from ApoE4's gain of toxic function rather than loss of protective function, so I'm somewhat optimistic about LX1020 but not as much their others. But it's early for them.
It seems like somatic gene-editing could kill a lot of birds with one stone here, but I don't have a great idea about how safe/cheap/reliable the delivery vectors would be at the moment.
Yes, but that "somatic gene editing" would need to be done on the glia, or perhaps neurons, within the brain. This is both difficult and dangerous. Also, I believe that tau misfolding is a prion disease, so once it happens, fixing the genes won't fix the problem (though it might slow it down a lot).
Point taken. I'm pretty leery about most applications of ML/AI, but I have to wonder if AlphaFold could design a protein to counteract prion folding?
My guess would be that's a problem at least as hard as "design a new laundry detergent to re-weave frayed fabrics," or "breed a new type of tree to proactively extinguish wildfires."
You can certainly find tree species that are highly wildfire-resistant, and there are such things as self-healing materials.
https://pmc.ncbi.nlm.nih.gov/articles/PMC10674826/
Obviously it's still a very hard problem, but... well, that's what we're trying to engineer superhuman intelligence for.
I would totally buy "Tide reWeave" laundry detergent, just to show off my freshly darned socks as a conversation starter.
Thank you for this excellent write up!
Wow!!! Thank you so much for this clear and cohesive breakdown of the theory and science underlying the disease.
Excellent write-up. Thanks for putting it together. Does anyone have more info on clearing the tau proteins directly and how that effort is going?
Thanks! I discuss where things are with targeting tau in the second half of this comment: https://www.astralcodexten.com/p/in-defense-of-the-amyloid-hypothesis/comment/146718972
Could I suggest adding a little summary at the end of the A->T->N pathway, and, in particular describing your argument that it's expected that anti-amyloid drugs have failed, given how they're used so far? (i.e. only after neurodegeneration has started). That was a surprisingly momentous/hopeful (and plausible) claim for being kind of just buried in the middle: that we basically already have a cure, and all that's needed to make it work is to scan everyone for amyloid buildup, so they can get the treatment a decade before symptoms would have started.
These extremely technical medical posts on here usually get the skimming treatment from me (and I suspect from a lot of other not-medically-literate readers) so ending summaries can be really helpful to reaching a wider audience! (I know you basically already have this under "The mechanistic claims", but that's still buried halfway through).
Also... I'm surprised this argument hasn't been put out in the press more. The total failure of anti-amyloid drugs seems like pretty damning evidence against the hypothesis, and such a plausible mechanism simultaneously explaining that failure and promising a cure seems like it would be an eyeball magnet for journalists.
This is a dense but highly readable analysis of the problem and I was surprised I stuck with it to the end. One thing I am curious about, is if AI can be of any help at leading to the proposed 75% solution mentioned under the testable prediction heading.
Neuroscientist working in an unrelated area, but I wanted to mention that a discussion of AD is not complete without mentioning the calcium hypothesis.
There is growing evidence that calcium dysregulation can drive both both amyloid aggregation and Tau cleavage, so there's the possibility that it may be upstream to both. Sleep disruption and TBIs also directly impact calcium homeostasis, so there's a mechanistic link.
See, for example:
https://www.sciencedirect.com/science/article/pii/S0021925819580760
https://www.sciencedirect.com/science/article/pii/S2090123225000268
Author makes two claims why ATN framing is preferred over the alternatives:
- tau hypothesis (TN) omits 'A' step, which is often precursor of 'T'
- infection hypothesis (IATN) adds unnecessary 'I' step, which is not directly causal with N and is sometimes unnecessary
I find this inconsistent.
Moreover, if the facts are layed out correctly (I have no idea), then TN/ATN/IATN are not really different hypothesis at all
They are still different:
1. ATN disagrees with IATN because it claims that there are ways to get too much amyloid without going an infection.
2. ATN disagrees with TN because it claims that there’s no way to get tau pathology without first having too much amyloid.
I want to join in and praise this excellent summary! I was also skeptical of the amyloid hypothesis from a much more superficial engagement with the field 10-15 years ago, and you have convinced me otherwise.
One thing that you have to learn in biology is ignoring outliers, and I think that was something that I got wrong.
You used the phrase "all instances" in your prediction, which is fine as rhetoric device, but is not quite how reality works. *Almost* all patients with a clinical Alzheimer diagnosis have amyloid plaques revealed by PET. Where "almost all" apparently means 88%. I think this made me severely suspicious when I was younger, because it means that 12% of the people with Alzheimer apparently did not get it via amyloid. But I have learned since then that in biology, 88% means "almost all", and the rest is a lizardmen fraction where something was weird. Perhaps the PET measurement wasn't accurate, perhaps the diagnosis was wrong, perhaps the patients did have some sort of exceptional Alzheimer which developed via some weird different pathway.
But it does make me skeptical of your prediction. Not about the direction, but 75% slowdown seems huge. Because not 100% of patients will react to the therapy, some will not respond for unfathomable reasons. And if 25% don't respond, you would have to slow down the disease by 100% in the rest, or more realistically, the therapy would have to revert symptoms in some patients. Which may happen, but which seems a strong prediction even in your bullish view.
It's true that in the past, Alzheimer's disease was diagnosed without biomarkers. Nowadays, the presence of amyloid pathology is considered part of the diagnostic criteria.
This doesn't make my mechanistic claim merely definitional or vacuous, because I framed it as a *counterfactual*: I'm saying that in all instances of Alzheimer's dementia (which I define in the following sentence as a whole collection of things including amyloid pathology), then counterfactually, amyloid-only therapy *would have* prevented dementia. (Of course, we can't directly test a counterfactual in individual cases, but that's why I made a separate prediction that can be tested.)
In an earlier draft, I had the parenthetical "(barring occasional instances of some other, separate and clearly distinct disease process, such as Parkinson’s disease dementia or a tauopathy with a different fold and different pattern of regional progression)". It was removed for brevity, but I still endorse it.
Ah ok, my number obviously doesn't make sense anymore if amyloid pathology has entered the diagnostic criteria. And thanks again! I still find the 75% in the testable prediction extremely optimistic, but it's not a fundamental disagreement. And time will resolve what the numbers are in the end.
Thanks very much for this article.
This is a statement not a question: obviously a common reason for a non-specialist to get very interested in a medical disorder is because they or a loved one suffer from it (or are likely to). I hope that’s not the case here; if it is, my best wishes.
> Severe enough amyloid pathology is a sufficient cause of Alzheimer dementia in almost all brains.
I haven't seriously investigated this issue, but I thought this was ruled out by the nun study? Or, like, they came up with the idea of 'cognitive reserve' that meant you could have lots of amyloid in your brain and not have Alzheimer's because you were losing brain you weren't using and still had enough left, but I thought it was pretty suspicious that the first time they looked at a significant number of controls the relationship got substantially weaker.
I discuss the existence of amyloid-positive, cognitively normal individuals in the article. The nun study (which I did not specifically mention) is one frequently-cited study which shows the existence of such people. This is predicted by the ATN model since it says amyloid does not directly cause neurodegeneration, but does so mediated by tau, after a ~15 year preclinical phase (I detail all the evidence for this model in the article). The question isn't whether such people exist, the question is whether even the ones with extremely high amyloid burden (not merely nominal positivity, since there are varying levels of susceptibility to the A->T conversion) ever go ~30 years without dementia.
In a PET study (https://academic.oup.com/jnen/article-abstract/70/10/832/2917208 figure 4), you can see that some cognitively normal people with "amyloid positivity" (> 23 centiloids) did not develop tau pathology within five years, but nearly all of those people were below 60 centiloids, and none of them were above 100 centiloids, which is approximately where I might draw the "sufficient amyloid" line. The claim isn't "all amyloid positivity results in Alzheimer's dementia", it's that "amyloid negativity prevents Alzheimer's dementia, *sufficient* amyloid burden basically guarantees it, and in between you have monotonically increasing risk".
(Overall, I don't find "cognitive reserve" to be a very useful concept: I think it's more an artifact of dementia and mild cognitive impairment diagnoses employing cutoffs on cognitive tests, not any actual sparing from an underlying disease process.)
Thanks for the reply!
Very well written and interesting post. I've seen my father drift off in his eighties. He did much much more headers than me, maybe I got knocked out once or twice more and I like my drink maybe even a bit more than him. I think it will be ok when my time of dementia will arrive. I don't have that much to lose.
Why is it easier/more common to target amyloid than tau? Amyloid seems worse since it's not the proximate cause of neuropathy and needs to be targeted or symptomatically.
Also, is there useful work on identifying the early amyloid plaques cheaply so we can target anti-amyloid drugs when they're most useful?
I think it’s mainly just that the importance of tau in mediating the A->N connection only became clear about 20 years after the causal evidence for amyloid itself, and drug development takes ~15 years from the first preclinical experiments, to early failures (such as N terminal tau antibodies), to trying more promising approaches (such as midregion antibodies and the IMO even more promising antisense oligonucleotides).
As for identifying amyloid pathology early: yes, this is possible to do today (via PET, CSF, blood, or plasma) but it’s not common because there haven’t yet been proven prevention methods. But once prophylactics are proven, I assume it’ll become widespread practice.
Very interesting article, thank you. I notice that I think it makes me a bit more confident than the amyloid hypothesis is correct compare to the other but on the meta level it makes me feel like a weathervane.
One question: I lost a grandparent to Alzheimer's. I am now worried about one of my parents. If I understood everything correctly we would want to start treatment something like 15-20 years before Alzheimer's start striking, which if the timeline are the same for my grandparent and my parent should be in the next decade or so. What can I do? What is possible to do?
CSF or plasma test for amyloid pathology, and enter a prevention trial (maybe trontinemab) if present.
Thank you!
What a great guest post! Thank you!
I remember a period after they found fraud in research supporting the role of soluble amyloid beta. There was a social media storm, and as far as I could tell nobody participating in it understood that this was about soluble amyloid beta, which is actually a different hypothesis than a role for amyloid beta plaques. Your source [123] seems to back up my memory. The subtitle is "A neuroscience image sleuth finds signs of fabrication in scores of Alzheimer’s articles, threatening a reigning theory of the disease", but the "theory of disease" seems to be a causal role for "toxic oligomers". Was this really "a reigning theory"? Can't there only be one reigning theory? And I would have thought that's plaques. Anyway, I think a lot of people casually following the story got mixed up.
There are many species of amyloid beta, varying in length, aggregation status, solubility, fold, and other post-translational modifications. Even though the causal role of amyloid has been apparent for a while now, there still hasn’t been a ton of clarity until possibly relatively recently about which species are “the toxic ones”, which ~20 years ago, before the mediating role of tau was as well understood, mainly meant directly causing neurodegeneration. Some preclinical evidence (including some of the fraudulent work) pointed to certain oligomeric species, but this was never a consensus, and was never the main reason people believed in the amyloid hypothesis.
Now we’re understanding 1) that the toxic effect is mostly mediated by tau, so we should be mainly looking for amyloid/tau interactions rather than direct neurotoxicity, and 2) that the harmful species are in greater states of aggregation and less soluble, such as plaques or protofibrils.
Superb write-up; thanks very much for this.
I would bet on the following: the leading* therapy in 12 years will not be one whose sole intended mechanism involves amyloid production or clearance, e.g. monoclonal antibodies, small molecules, or gene editing that work directly on amyloid.
“Leading” means some combination of efficacious and most widely used.
This is the important sense in which the amyloid hypothesis is wrong: it may be “correct” in a narrow biochemical sense, but it is not helpful.
(edited)
@grok read all the footnotes and tell me how to cure alzheimer's
searching "Elon Musk opinion how to cure Alzheimer's"
When did those scientists realize that they need to treat amyloid BEFORE it's too late? Recently? Otherwise why were there those late treatment experiments? I guess we do realize it recently, maybe only because of those failed late treatments?
I think the consensus gradually emerged over the last decade, mainly due to:
1) The increasingly clear importance of intermediate pathologies like tau which mediate the effects of amyloid.
2) The modest clinical benefit of drugs like lecanemab, and donanemab, despite significant plaque clearance.
A challenge when conducting clinical trials is you ideally want to test in a population which is likely to decline in the course of the trial, and for the trial to not run too long, so as to maximize statistical power, minimize cost, and maximize speed of discovery. So until it became very clear that earlier treatment was vital for amyloid therapies, there were natural pressures pushing in the opposite direction.
Question about the clinical trials of lecanemab, donanemab, aducanumab. I understood that the clinical trials calculated the change in the speed of progression by looking calculating (cognitive score at T2 - cognitive sore at T1) / (time between T1 and T2). This doesn't rule out the effect being to simply improve cognition by a fixed amount, without impacting progression. Is my reading correct, or did I miss something? Asked differently, is the improvement from these drugs different in kind from the improvement from memantine or donepezil?
An important difference compared to donepezil (and other acetylcholinesterase inhibitors) and memantine is that 1) the advantage over placebo plateaus after a few months, and then your trajectory of decline matches placebo but vertically shifted and 2) when you go off those drugs, you revert to the trajectory of the placebo group.
Whereas when you go off lecanemab, you maintain your advantage over placebo or late starters of lecanemab: https://www.alzforum.org/news/conference-coverage/leqembi-case-long-term-dosing
And for both lecanemab and donanemab, the advantage over controls widens over time: https://www.alzforum.org/news/conference-coverage/signs-lasting-benefit-amyloid-immunotherapy
Thank you, that explains it well.
I don't have any relevant expertise or knowledge, so the following probably just reflects my ignorance. But if we stipulate the ATN model as OP explains it, it seems to me that there are daunting problems with targeting amyloid (and so I don't understand why OP is so optimistic about it).
If treatment really has to start 10-20 years before symptoms to be fully effective (because tau degeneration is self sustaining once kicked off) then to significantly impact the disease (a) you would need to aggressively screen and treat relatively young and unsymptomatic people, and (b) proving that the benefits of doing so outweigh the risks will take decades. Is society really going to decide to start giving every 50 year old a PET scan and then putting some significant fraction of them on monoclonal antibody infusions for decades (that might have significant side effects) based on extrapolation from 18 month studies in much older, symptomatic patients that (can at most) show small, statistically signficant but clinically insignificant benefits? Is anyone even going to try to run such a difficult randomized trial over the course of 15-20 years, in hopes that the results will help people in 35-50 years? Treatment that attacked a later stage of the A->T->N pipeline would have a huge advantage in timeline for both testing and deployment.
What am I missing?
Thankfully, there are plasma tests that have recently been validated and aren't too burdensome or expensive to administer.
As for results from prevention trials, we should get results for donanemab's in 2 years, and lecanemab and trontinemab in a small number of years after that. While that trial may not involve treatment quite as early as I'm suggesting, if the earlier = better principle is correct, we should hopefully start to see substantial improvements in efficacy compared to current approaches, and I'd imagine there'd be a shift towards treating as early as possible.
The main thing is that the first people who get prophylaxis or treatment in the asymptomatic phase of the pathology will be people with extremely high risk or even the AD or chromosomal forms of Alzheimer's. They will be eager for treatment and the risk/benefit will look better for them. Data from those people will inform management for patients with lower risk.
There isn't enough money for widespread treatment at $30k/dose; my expectation is that ASI/nanotech will soon reach the point where everyone is either cheaply cured and on UBI, or dead.
>This fraud has sometimes been framed by amyloid hypothesis critics as impacting very foundational work for the amyloid hypothesis. But this mostly isn’t true.
Excellent summary! After the fraudulent research was exposed, I wasn't sure where we stood with the Amyloid theory. Of course, science journalists are mostly ignorant of the subjects they write about, and reliance on popular articles will give you a distorted view of the questions. I never had the time or the motivation to dig into the literature. Thanks for doing what the science journalists should have been doing. :-)
I know nothing about this topic but I felt an overwhelming urge to come here to plug the book Valley of Forgetting by Jennie Erin Smith, about the lives of a group of Colombian families with autosomal dominant PSEN1 mutations (as mentioned in the post) and the scientists who have been studying them for the past few decades. It's a really good book. That's all.
Thank you for this. I am confused about your bet. If amyloid build up typically predates taopathis by 15yrs and amyloid alone doesnt cause much cognitive decline, why would you expect studies targetting amyloid build up to be completed within 12 yrs? They have already begun?
Some have begun; and also plausibly one only needs to clear amyloid pathology prior to significant tau pathology, not at the very beginning of amyloid positivity; and also my bet is only aiming for 75% efficacy (I’d expect more over time).
What do people think about the claims that Ozempic is protective against Alzheimers/dementia?
So far, our evidence comes from non-randomized studies of people with diabetes (comparing those on Ozempic to those on insulin). There are parallel phase 3 RCTs due to finish next month, in people with Alzheimer's disease, which will be a better test of the theory.
Is it just me, or is there a slight bait and switch going on here?
This is not a topic I'm that informed about. But my simple understanding of the amyloid hypothesis was indeed that amyloid is directly causing the symptoms, which considering all the early antibodies and mouse models seems to also have been what researchers though; Wikipedia still has the amyloid hypothesis as amyloid directly causes the symptoms without mediation. And the standard arguments against the amyloid hypothesis likewise focus on the bad correlation between amyloid plaques and impairment.
From a cursory search, it indeed looks like currently, amyloid hypothesis in the research is mostly used in the amyloid tau cascade model sense you describe. Which does sound more convincing than the original hypothesis. I don't know if this is simply a failure of communication, but it seems like what the amyloid hypothesis refers to has quietly changed in the last decade or two, without that percolating in the regular critical discourse about it. I'm undecided as to whether it has changed so much that it should really be considered a different hypothesis that should be given a new, distinct name. But clearly, the amyloid hypothesis as stated twenty years ago is false even according to your model of the disease.
I think this is fair! This seems analogous to how "atomic theory" turned out correct but it was later discovered that what had already been called "atoms" weren't the atomic units after all. Is this a bait and switch? Well, maybe! I don't know enough about the history there, but I imagine it was a combination of 1) some terms undergoing a natural scientific evolution as our understanding improved, 2) some scientists protecting their egos, and 3) the challenges of communicating science to the general public, who have a very low tolerance for complexity. I think there's probably a similar breakdown in this case.
I only mean to defend the ATN model, and more generally the idea that amyloid plays a necessary and (in sufficient severity) sufficient role, not the idea of proximate neurotoxicity.
Do you know which model was the impetus for the current antibodies in the pipeline? If amyloid is the ultimate cause it's probably a good target for prevention. But if the actual toxic effect goes through tau that would seem to be the obvious target to treat people post symptom onset (instead of decades beforehand). I'm not aware of any tau antibodies currently being tested, is it fair to assume that so far, all treatments that made it to clinical studies did indeed assume that amyloid is directly causing the symptoms?
To your first question, depends on how you define "in the pipeline". The amyloid antibodies currently on the market began development as far back as 2007, in particular lecanemab's initial murine form "mAb158" [1], motivated by the so-called "Arctic" Alzheimer's disease mutation, which was known to result in more amyloid-β protofibrils but not actually yield plaques. So the theory there is that it's the protofibrils (rather than the less-aggregated species like monomers and dimers) but not plaques *per se* which were the problem.
In that paper, they say that "oligomeric Aβ, including the protofibril, is thought to be a primary neurotoxic agent in AD", which was a commonly held view at the time. So that would indicate a belief in a fairly direct neurodegenerative effect.
Donanemab development began in 2012 in mice as "mE8" [2], motivated by a perceived need to target plaques rather than less aggregated, soluble Aβ (which is slightly different from but reconcilable with the lecanemab rationale). They suggest a more indirect viewpoint, saying "amyloid deposition is the causative factor for the initiation of the neurodegeneration cascade, which includes inflammation, gliosis, neuronal damage, synaptic loss, and cell loss", though they don't mention tau specifically.
By now, virtually everyone who believes in the amyloid hypothesis, including those working on e.g. trontinemab, accepts that most neurodegeneration is mediated by tau pathology (or something closely connected to tau pathology).
Most clinical trials since the year 2020 (but not before) have had targets other than amyloid (see my references [126, 127] in the essay), with tau probably the most common non-amyloid target. Four antibodies (semorinemab, gosuranemab, zagotenemab, and tilavonemab) failed, but we now have much better structural understanding from cryo-EM which suggests these antibodies were targeting the wrong part of the tau protein to prevent the prion-like misfolding.
Newer antibodies such as bepranemab (UCB0107), posdinemab (JNJ-63733657), etalanetug (E2814), and BMS-986446 target tau's midregion and have a better chance of success. bepranemab is the only to have completed phase 2 so far (posdinemab should be next, in about six months). It slowed tau accumulation by ~half but failed on its primary cognitive endpoint, although several secondary cognitive endpoints showed indications of benefit, and in a pre-specified subgroup analysis in a "low tau" population (so the very early symptomatic stage, when tau pathology is just getting started), this signal was a bit stronger.
One explanation for this may have to do with tau pathology worsening by two basic mechanisms: intercellular spread and intracellular replication. Antibodies mainly act extracellularly, which can help to slow spreading in the early stages, but once there has been enough spreading, most worsening is occurring intra-cellularly and requires a different strategy. This is like how early in a pandemic shutting down air travel is helpful, but once there's a little bit of virus in every region, local efforts are going to be necessary too.
A strategy consistent with this hypothesis is antisense therapy, such as BIIB080 (MAPTRx), which suppresses expression of the tau gene to provide a smaller "susceptible population" for the prion-like replication. In its phase 1b trial, it didn't just slow tau pathology — it actually *reduced* tau pathology, which is the first time this has been achieved in humans. And while not powered to detect cognitive benefits, it appears to have slowed actual cognitive decline by about 2/3 (with very wide error bars) vs. controls from a different study. So I think this is the most exciting approach for targeting tau. Phase 2 began last year.
[1] https://onlinelibrary.wiley.com/doi/abs/10.1111/j.1471-4159.2007.04759.x
[2] https://www.cell.com/neuron/fulltext/S0896-6273(12)00950-6
So I guess I should withdraw my bait and switch accusation, although the phrasing is definitely confused in the literature. From a 2016 paper on aducanumab:
> The amyloid hypothesis posits that Aβ-related toxicity is the primary cause of synaptic dysfunction and subsequent neurodegeneration that underlies the progression characteristic of AD. [1]
Which does kind of sound like Aβ is directly killing your neurons to me. But then they reference a 2002 paper for that claim, which seems much more carefully phrased and aligns well with your argument, even explicitly mentioning tau:
> According to the amyloid hypothesis, accumulation of Aβ in the brain is the primary influence driving AD pathogenesis. The rest of the disease process, including formation of neurofibrillary tangles containing tau protein, is proposed to result from an imbalance between Aβ production and Aβ clearance. [2]
So maybe all the focus on Aβ antibodies just merged that in the mind of most non-domain experts talking about the topic (and maybe also some domain experts). Anyway, thank you for restoring some of my faith in medical reserach (or this particular field of it). Apparently part of the issue here was just the long time between initial development and clinical trials, and the field is indeed capable of eventually changing course once it becomes clear that the current approach isn't yielding any results.
[1] https://www.nature.com/articles/nature19323
[2] https://www.science.org/doi/10.1126/science.1072994
> After it forms, the tau pathology no longer appears to require amyloid’s assistance to keep spreading (although amyloid may still accelerate it). This probably explains why existing anti-amyloid therapies have been only ∼30% effective in test patients, who are usually late in the amyloid → tau progression even if early in having symptomatic disease.
This would seem to suggest that anti-amyloid therapies _would_ work, if we gave them to people who were going to develop Alzheimer's disease in the future rather than giving them to people who have already developed Alzheimer's disease in the past.
The post suggests that many such people are easy to identify. What does this experiment show?
We’ll have our first data from plaque-clearing antibodies in 2027, when donanemab’s prevention trial completes.
Overall, this is a truly excellent piece of exposition. Thank you. I think the "frustration" section is unnecessary and possibly detracts from the piece as a whole.
What do you think about this 4-part series on an apparently influential paper (Games et al.) that isn't cited above? https://journalclubwithmyka.substack.com/p/of-mice-and-mechanisms
Parts 2, 3, and 4:
https://journalclubwithmyka.substack.com/p/of-mice-and-mechanisms-part-2
https://journalclubwithmyka.substack.com/p/of-mice-and-mechanism-part-3
https://journalclubwithmyka.substack.com/p/of-mice-and-mechanism-part-4
I'm worried that it often takes this level of 1) training/knowledge, and 2) willingness to spend 10s of hours on a single paper, to see the actual problems.
I think it's right to criticize the old highly over-expressed amyloid-only mouse model for not being a full recapitulation of the disease (doesn't have tau neurofibrillary tangles). But it:
1. Doesn't engage with all (or really any) of the extensive evidence I discuss, from amyloid+tau mouse models and humans, for amyloid → tau causation. In the human disease, tau pathology has an origin independent of amyloid (so isn't captured by the amyloid-only mouse model they focus on), yet amyloid is also necessary (and in enough severity) sufficient for that pathology to extend beyond the medial temporal lobe (MTL), and usually for it to locally worsen within the MTL.
2. Doesn't offer an alternative hypothesis to explain the genetic evidence for amyloid's causal role.
3. For antibodies which do clear plaques, makes the misleading "∼0.5 points on an 18-point CDR-SB scale" argument I refer to about antibody efficacy (a perfect drug could only have achieved ~1.5 points efficacy). Doesn't engage with the open-label extension evidence of widening benefits after the initial 18 months of treatment.
4. For treatment before cognitive symptoms which I argue for, relies on the clinical failure of amyloid antibodies such as solanezumab which also failed to clear plaques. (The ones which succeed at clearing plaques haven't finished the relevant clinical trials.)
5. Frames the (I agree very bad) fraud as pertaining to "core Alzheimer’s findings that had guided decades of research and many hundreds of millions in funding", which isn't really true, as I discuss.
6. Misunderstands the lower soluble Aβ42 in most Alzheimer patients' cerebrospinal fluid, saying it's a new finding that challenges the amyloid hypothesis. But we know that it's due to impaired clearance (the missing Aβ shows up in amyloid plaques on PET scans), and in fact for this reason it's been considered a standard biomarker of amyloid pathology for decades.
I shared this with an expert in the field as I have not kept up with the literature since retirement in 2022. Her comment was: "He actually did a great job distilling a complex picture and history into a very current and comprehensive view, and provided a nice bibliography to go along with it." It was interesting to see the exciting preliminary results from the Trontinemab trial. You have positively changed my bias point about future Alzheimer's treatments and provided a very readable example of the way science slowly, but positively progresses. Thank you for this excellent writeup that has helped to bring your readers up-to-date.
It’s impressive the author has made a specific and time-stamped prediction about the progress of clinical trials in this field. So often people “predict” without standing up for it.
Ultimately, biological plausibility is great but it is no substitute for clinical outcome trial evidence. So for now, it remains a hypothesis awaiting confirmation.
Off topic, but what does safety mean in terms of AI?
This term is used by different people in different ways, but on this blog, the usual concern about “AI safety” is “how can we make sure that we don’t end up with an AI that is so unsafe that it leads to human extinction?” This term has been partly overloaded for lesser things, like “make sure AI does not use naughty words”, so some people have looked for terms that are less likely to be “stolen”. The current leading candidate is notkilleveryoneism.
Thank you!
You’re welcome! There were some posts on the old blog that talked a bit about this; I feel like almost a quarter of those tagged “ai” at least mentioned it: https://slatestarcodex.com/tag/ai/
The new blog has some too but they’re harder to find because substack doesn’t have tagging. (To be fair I’m not sure Scott was perfectly consistent in tagging on Wordpress but still.)
I've suspected for years that I'm having mild cognitive decline, but I'm under the impression that there is no affordable test that would test positive years before I fail any cognitive test. And since I'm "too young" to get Alzheimer's, I expect no doctor would order an expensive or invasive test. Given that companies are still pursuing Aβ treatments, is there a good, safe, affordable test in the pipeline? If someone is very interested in getting a test, is there some plausible way to get one?
P.S. twice now I've experienced sudden partial vision loss, which I interpreted as a potential stroke (both eyes affected equally; parts of visual field replaced by "snow"; fovea unaffected; the "snow area" slowly changed shape; problem disappeared within an hour). Before that, I woke up one day and noticed that some sharp corners on fonts and shapes looked rounded, although I looked closely to confirm that they were, in fact, sharp corners. I figure if my visual system is suffering minor damage, the rest of my brain probably is too. These phenomena seem like mini-strokes rather than Alzheimer's, but I don't know anything I can do for diagnosis or treatment of mini-strokes either. My doctor seemed unconcerned, so nothing was done.
A p-tau217 blood test (Quest Diagnostics) is $1264.00 at Ultalabtest.com. My results were back in two days. There were multiple phlebotomy sites to choose from.
Thanks. From what I can tell by a quick search, they make no representation about how early in the progression of AD the test should turn positive or "indeterminate". Not surprising for a non-FDA-approved test, but for that price I'd need a test that reliably detects AD very early.
David, for your sake please contact a neurologist and get yourself screened to rule out multiple sclerosis. The optic nerve is commonly one of the first to be affected in a way that the patient notices.
Among the many early signs of MS are the vision problems "optic neuritis, double vision (diplopia) involuntary eye movements (nystagmus)". I don't have these symptoms.
That was a truly amazing post. Well written, well reasoned, and well sourced. I followed the link in footnote [124] and was amazed again, this time with horror. I hadn't realized it was that bad.
This is some of the best science writing I have ever read.
Very useful and well written article, thanks a lot!
Thank you. The fire analogy really clarifies the stakes and the timeline. The scary part is when you map the 10-20 year preclinical window onto demographics. It means the fuse for the baby boomer retirement crisis was lit around 2005, and we're now entering the critical decade to act before the house really starts burning down. A "Warp Speed for Alzheimer's" similar to the deployment of the COVID vaccine feels like the obvious response, because the alternative is letting it consume ~$1 trillion/year (per the Alzheimer's Association's projections), tying a lot of work force and loosing endless DALYs. I wonder which country will be the first to employ a concerted effort. I feel like my own country (Germany) might be better equipped than the U.S. currently. Feels like the first country to treat this as a national industrial strategy rather than a healthcare cost is going to win big.
Missing reference to Bredesen, FINGER, MIND diet. Amyloid hypothesis is not defensible. There is resistance to any alternative IMO because establishment researchers knew the faked studies and doubled down for ego defense. Establishment researchers who knew and cared about science, left the establishment and have been going at prevention for years and are now ahead by years. Acceptance of their peer reviewed research by the establishment amyloid researchers would be a blow to egos.
I don't understand why some people call Alzheimer's type 3 diabetes. Perhaps Alzheimer's is growing like type 2 diabetes epidemic?
Is Alzheimer's also related to metabolic disturbances?
Perhaps the diet of starch, sugar and PUFA has contributed to Alzeimer's epidemic?
What I find particularly striking about this whole discussion is how much it reveals about the huge importance of social dynamics in the sciences.
It took you, a very smart and highly educated person, quite a long time to dig through these papers and even then you are highly reliant on the arguments and opinions about this area willing to be circulated in public.
I'm guessing you do have a pretty good picture of the state of the research but it wouldn't take much to shift the plaque hypothesis from disfavored but talked about to too embarrassing to mention in academic circles (eg the way geologists refused to consider a huge flood as the cause of the badlands as it felt too much like discredited biblical based geology) and you would have been unable to gather much supporting evidence.
And beyond this my sense is that a huge amount of the knowledge in a field is never exposed in published articles but merely conveyed by a sense amoung those who do the experiments of what does and doesn't seem to work. So we really are, probably unavoidably, extremely reliant on the social dynamics of experts in the field.
And no doubt this is why even sophisticated investors like drug companies and VCs find it unprofitable to fund many outsider ideas. I mean funding a view that's seen as less likely but still serious is one thing but the second you fall off that cliff it becomes almost impossible for anyone who isn't directly involved in the field to tell the difference between something that's a 1 in 100 chance of being true and 1 in a million.