Since the recent disappointingly small effects of monoclonal antibodies on Alzheimer's Disease (AD) progression, there has been lots of discourse around what the cause of Alzheimer's might be. "If not amyloid then what is it?", many wonder. What is the thing we have to remove? In cancer we remove cancer cells, in treating cardiovascular disease we aim to lower LDL particles and that massively lowers risk.

If not amyloid, what is the LDL of Alzheimer's?

In this post I argue that there is no answer to that question: instead it's more useful to think of AD causes as multifactorial where each doesn't necessarily guarantee an AD diagnosis in every situation but it increases the odds of it all else equal. And moreover and perhaps most importantly I argue that is useful to think separately about what to do to prevent Alzheimer's (easy, address risk factors) and what to do to reverse or stop Alzheimer's (harder, requires intervening upstream).

I currently work at a company (Retro Biosciences) that has AD among its indications, but my interest in this disease predates Retro; before I joined I wrote a few posts about it you can find here. I found Alzheimer's quite intriguing because cardiovascular disease seemed simple: LDL cholesterol clogs arteries, lowering LDL lowers disease risk. But with Alzheimer's that didn't happen with amyloid beta and everyone was at the time talking about pharma being misled by data falsification or by intellectual stubbornness around the amyloid hypothesis. What was going on? I wondered back then and out of my wondering came those blogposts. On the amyloid hypothesis a recent post worth reading to steelman it is this one (The autor predicts clearing amyloid will eventually work, I predict it won't).

Perhaps as a result of having mainlined a good deal of philosophy for many years, I can't help but start with examining what we mean by "cause" and what we mean by "Alzheimer's".

I wrote an introduction to what Alzheimer's is here; but briefly Alzheimer's is a disease the onset of which is defined by elevated levels of amyloid beta in the brain, as measured by a series of agreed-on proxy biomarkers (Alzheimer's Association Workgroup, 2024). Following the appearance of amyloid one then gets tau tangles and neurodegeneration. If an individual gets tau tangles but no amyloid, that is classified as PART (Primary Age-related tauopathy) or other non-AD diagnosis. This definition does not depend on cognitive symptoms, so someone that has their brain full of amyloid but is cognitive normal will still be considered to have Alzheimer's by this definition.

Definitions are meant to serve some useful purpose, they are not by any means "objective" (Something that has been obvious to me for over a decade but that sometimes people have fights over). We are free to define as we wish! I say this because some may accuse the AA Workgroup definition of being amyloid-biased, privileging the role of amyloid in the disease, implying that AD is "about amyloid by definition". We may wish to define AD in some other way, but I personally like this definition: there are many reasons why people lose cognitive abilities as they age (another one is Parkinson's) and it seems like the most common such way has to do with amyloid: that's what is found in the brains of most of those patients, so having a label for "the process that tends to go along with presence of amyloid in brains that tends to lead to neurodegeneration" is useful.

Now I suggest reading the post I linked above about the failure of monoclonal antibody therapy: there are therapies out there that are able to clean up the brain (as measured indirectly via plasma or more directly via PET scan) of amyloid and yet these patients still experience neurodegeneration. Does this mean the patients are cured of Alzheimer's and yet continue to decline? The document kind of skirts around this issue by saying that the purpose of the definition is defining the disease based on the natural course of the disease, not to be used for diagnosing post-treatment. They authors do admit that the underlying process continues active despite the amyloid clearance.

This is a limitation of their definition: what is that mysterious underlying process?!

Imagine a simple model where we have two things: Alzheimer's, a disease denoted by amyloid presence, and then something we could call "secondary neuroinflammatory dementia", another disease caused by Alzheimer's.

This would be similar to having diabetes and then having the diabetes causing diabetic foot. In that case you could treat the process that is making you lose the foot without necessarily resolving the diabetes first. Similarly one could stop AD-caused neurodegeneration without curing Alzheimer's. There are patients with amyloid but no cognitive decline, so they have Alzheimer's, but that's then made irrelevant as their cognition does not worsen. Ideally we would want to cure both and leave the brains squeaky-clean of amyloid too, but in the timeframe that is of relevance now, having a direct cure that makes patient function better seems important.

That's it for the definition of Alzheimer's. Let's now talk about causes. What is a cause? People use the word in different ways and I'm not going to try to go into formally defining cause as that's a fraught topic in philosophy. Instead I'll use an assortment of definitions that people might use. A cause is...

1. A sufficient reason why something happens
2. A sufficient and necessary reason why something happens
3. A factor such that if present, it increases the probability of the thing happening (aka a risk factor)
4. A molecular entity such that if removed, the symptoms of Alzheimer's are substantially ameliorated

Here's why AD doesn't fit neatly with this:

1. There is no one phenotype that guarantees Alzheimer's. Even for very aggressive familial early onset mutations, one needs to add aging as a secondary factor.
2. There are no such reasons to be found. Aging is a necessary reason but it is not sufficient, at least in the timespans we have thus observed.
3. There are indeed many such factors. I'd prefer calling these risk factors to keep in mind that they are not deterministic. Cause sounds very deterministic!
4. Amyloid beta could have turned out to be this, but it hasn't thus far. Some hope tau will be this but so far trials have been lackluster,

A very simple understanding of what goes on in Alzheimer's is that for a variety of reasons (Different in each person), the clearance and production of amyloid beta is unbalanced leading to its accumulation in various forms; in turn the presence of amyloid can beget more amyloid because the inflammation they cause can downregulate the very mechanisms cells use to clear it up. That then leads to tau tangles and neurodegeneration. As people age, cells become more trigger-happy with inflammation and less resilient overall so by the time a brain is aged, it can get locked in an equilibrium where cells are dying for no good reason.

It is because there's a variety of reasons that can trigger this imbalance that I say that Alzheimer's has no cause. The following things (the list is not meant to be exhaustive) can increase the odds of having Alzheimer's:

• Mutations in the genes APP, PSEN1, PSEN2 (near certainty of getting AD, as early as 21 years old). Though even these are not "necessary and sufficient" for the classic disease involving neurodegeneration as: a) One has to still age b) In the presence of other mutations like Christchurch one does not experience cognitive decline (Arboleda-Velasquez et al. 2019).
• APOE4 (Fortea et al. 2024) is also another notorious gene variant that contributes to highly increased odds of AD along with a long tail of other genes that show up in GWAS like TREM2.
• Traumatic Brain Injury is associated with increased odds of diagnosis (Graham et al. 2022; Zhang et al. 2021; Mavroudis et al. 2024). Worth noting that TBI itself causes amyloid and tau to appear transitorily in the affected area.
• Gingivitis is also possibly associated with risk of AD (Mo et al. 2025) perhaps via Porphyromonas gingivalis (Singhrao et al. 2015; Sarmiento-Ordóñez et al. 2025). This was the basis of the company Cortexyme: addressing AD through curing gingivitis, but their trial didn't meet their primary endpoints and was discontinued.
• Herpes Zoster (varicella) presence also contributes to AD, which is why being vaccinated lowers diagnoses by around 20% (Xie et al. 2025)
• Unsurprisingly, so does HSV (Araya et al., 2025; Itzhaki et al. 2021) though trying to stop the virus once one already had AD fails
• Down syndrome (because of an extra copy of the APP gene in the chromosome that has an extra copy)
• Aging

One could call all of these "causes of Alzheimer's" if one wants. I don't know about you but if I hear "gingivitis causes Alzheimer's" that sounds to me like "If you have gingivitis you will have Alzheimer's" which is not true. When crafting and using definitions and names the way it tends to make one think and the meaning it tends to convey matters. In a context with well defined terminology (Like Judea Pearl's terminology for causality) one can use those terms just fine, but if you suspect using "cause" will imply in your listeners or readers "deterministic" then don't use that word unless you mean it.

But I care more about what we can learn from these being contributors to AD in the first place (Instead of debating what a cause is): some seem to have to do with amyloid itself and others have to do with neuroinflammation.

That is, I think we shouldn't go and say "Cause X is associated with AD, therefore let's treat cause of Alzheimer's X" rather I say that it's more productive to say "let's use the associations of XYZ risk factors with AD to build a model of AD and then use the model to decide where to intervene".

The former mode of thinking is very common in biology because biology is hard and going one thing at a time, or testing things that are easy to test first makes sense: Take Alzheimer's incidence, throw a bunch of variables in a regression, see what the risk factors are, see if fixing any of that cures it one by one. I think this approach is fine for Alzheimer's prevention. Alzheimer's prevention is easy: All the risk factors I listed above can be currently avoided (except aging), but the risk can be brought from large (familial mutation or ApoE4) to manageable (<5% lifetime). In contrast treating Alzheimer's is a more difficult problem and it is this that I think the most about as it remains an unsolved problem.

What do we learn by looking at these risk factors that can be useful for treatment? Why would HSV or gingivitis for example increase the odds of Alzheimer's? One easy guess is that the presence of a virus in the brain leads to a response against the virus: inflammation. Inflammation in turn triggers the production of amyloid beta which seems to have some function as an antimicrobial peptide (Gosztyla et al. 2018). Inflammation however is not an surgically precise process: it's not like "if and only if there's virus then produce Abeta", inflammation can be a very general response and this gets disregulated with aging to the point where cells' baseline level of inflammation (proxied by overabundance of AP1 and Nf-kB motifs in open chromatin) is higher.

This latter fact is part of why it is difficult to treat Alzheimer's disease in humans compared to Alzheimer's mouse models: once the positive feedback loop where inflammation begets amyloid and amyloid begets inflammation starts, it doesn't seem to matter that much that one removes the original reason why the cells got inflamed in the first place: that epigenetic memory of the inflammatory trigger is now etched in the chromatin: this is why aged cells, when faced with an inflammatory cue like LPS tend to secrete more inflammatory cytokines (Frank et al. 2010; Moden et al. 2012). Of course, aging also makes the cells less functional in general, so modulating inflammation alone may not be enough.

With this knowledge what to do becomes different: instead of removing the triggers, we could, for example, do some of the things we are doing at Retro:

• Make neurons more resilient (for example by boosting autophagy; those patients that have plenty of amyloid but no neurodegeneration have increased autophagy (Tumurbaatar et al., 2023). Could we make more patients artificially be like these pharmacologically?
• Replace cells with freshly made ones that are more functional and less pro-inflammatory (Rao et al. 2025)

This is not what most of the field has historically done. Most of the field was for a long time enamored with there being "a" cause for AD, in particular a thing, a molecular entity you can remove that will stop the disease altogether (a species of amyloid or tau), and indeed to this day tau+amyloid are the dominant class of drugs in the pipeline to treat AD (Cummings et al. 2024). But as you can see from the linked source, biopharma is now exploring many other alternatives.

In conclusion, the evidence we have from human observational and interventional studies as well as genetics suggests, in my view, that to treat (not prevent) Alzheimer's it is best to forget about these risk factors and use the data instead to model what one observes and derive treatments from the model.

Attempting to modify the risk factors of this disease once it has already started is unlikely to lead to cures. In a tweet that motivated this post, I said that "Alzheimer's has no cause". Ultimately this is the point that I wanted to make in that tweet: that we should stop thinking about these various contributing factors that much and instead think more deeply about the underlying dynamics of the disease (autophagy, inflammation, aging, etc) and derive interventions from that.

In my view, targeting those processes is more likely to lead to Alzheimer's cures.