Molecules won't defeat aging
Why AION is choosing fields
Point blank: a molecule is not going to solve aging. Here are some thoughts on why.
At AION, we're working to bring aging under complete biomedical control. As a heuristic, we've started measuring every decision we make in the lab against whether the approaches we're using will translate to this goal state.
A few months ago, this led us to drop molecular approaches entirely. That includes drugs and gene therapies. When I mention this to people used to traditional methods, it sounds radical because we’re essentially ruling out the core paths everyone else is chasing. To us, it's just common sense.
Let’s assume there is a single drug out there that could lead to something dramatic like a 50%+ lifespan extension outcome. We can infer the probability of this compound existing by looking at the DrugAge database, a single source that aggregates data from thousands of lifespan studies across species. Mus musculus (aka mice) is the organism with studies in the database most closely related to humans. Pulling these results, we find 134 different compounds tested with only 103 having been replicated more than once for reliability.
Let’s give the benefit of the doubt though and look at the full set of 134, where the mean average median lifespan change clocks in at 4.04%, the median at 2.8%, and the standard deviation at 6.77%. For maximum lifespan across the 111 compounds with complete data, the mean hits 2.56%, median 1%, standard deviation 5.26%. Rats show similar modesty with 23 entries averaging 4-5% extensions. Replication kills the hype, initial effects shrink in multi-assay averages. Fitting a normal distribution to the average median changes yields a 5.34e-12 probability of >=50% extension, meaning 187 billion molecules screened to expect one hit1.
At $500,000 per compound for replicated mouse longevity assays benchmarked from NIA Interventions Testing Program protocols (200-300 mice over 3 years), that totals $93.5 quadrillion. Oh, and add a $1 billion for the lone projected success case to validate in humans.
In lieu of a similar analysis for gene therapies, I will talk about challenges with delivery for them below.
A bit of background: there are essentially three approaches to tackling the contributors to aging.
Removing accumulated damage.
Replacing organs or the whole body.
Reprogramming aged tissue to a more youthful phenotype.
We've hung our hat on approach 3 for a few reasons.
Adding to the comments above in regard to path 1, we figure the effort to develop ways to address the full set of accumulated damage from aging, things like stiffening of the extracellular matrix, ROS buildup, senescent cell accumulation, mitochondrial dysfunction, and more, would require so many concurrent interventions that it's smarter to focus on upstream causes instead. Plus, the current pharmaceutical landscape has this angle covered. The race in drug discovery to target specific indications will eventually produce drugs for this net set of aging byproducts.
As for 2, while scientific hurdles remain, it's arguably a surgical engineering problem now. We know organ replacements work individually, and multiple companies are now exploring knocking out the forebrain in human clones to grow mature bodies for organ harvesting or using the whole thing as a replacement vehicle. We’re not pursuing this direction both because other players have this space covered and the approach has bad aesthetics. I have a pet hypothesis that the universe rewards pursuits with good aesthetics.
Now for 3: Since 2006, biologists have been able to turn any cell type into a stem cell, proving we can control a cell's state. The logical extension is that if we can revert a differentiated cell to a stem cell, maybe we can scale that to rejuvenate an aged organ or even a whole human.
The biggest longevity companies agree. Altos, Calico, Retro, New Limit, and others have raised billions to chase reprogramming for tissue rejuvenation. That said, their current molecular approaches hit fundamental limits in spatial and temporal control.
The most successful in vivo reprogramming study relied on transgenic mice with a doxycycline-inducible switch for Yamanaka factors (OSKM). They turned it on systemically but only in short cycles to nudge fast-responding tissues like pancreas and muscle toward partial rejuvenation, while avoiding teratomas or over-reprogramming in slower spots like neurons. No organ-specific tweaks needed, but the gains stayed limited to the quick-recovery systems because if cycled too long, you risk tumors or other cell chaos. Unless we identify a way to edit our own genomes to have this switch, this approach only works for future generations that would choose to engineer their embryos in such a way.
Even if we cracked universal adult insertion, the mismatched reprogramming speeds across tissues (liver beats kidney, kidney beats skin, skin beats neurons with plenty of gradations in between) would force custom gene circuits for each cell type to dial in timing and strength, all while dealing with natural shifts like differentiation, dedifferentiation, and transdifferentiation that make genomic fixes fade over time. Episomal vectors for non-integrating factors might dodge the permanence issue, but you’d still need perfect delivery everywhere which would be tough with AAV limits on tissue reach and immune blowback, not to mention the need for perfect pharmacokinetics in each patient to trigger it via an activation drug without off-target effects.
Why bother solving delivery challenges for each subsystem, where timing the expression of reprogramming factors is a nightmare, when we could achieve external spatial control instead?
The only way to do that is with fields.
Some deeper background on why we at AION went all-in here: after identifying the basic bioelectric reprogramming signature tied to cell reprogramming, we started thinking about how to impose it on all cells in our experiments for higher efficiency. Problem was that the information we had to go off of was (X) chemical acts on (Y) mechanism to change membrane potential in a positive or negative direction. That said, the level of fidelity of information we required for our work to continue was: (A) concentration of (X) chemical in (B) cell type leads to (C) degree of change(eg, 5mV, 20mV, 50mV) over (T) time, which would require a non-trivial amount of chemical screening and moreover developing a microfluidic system for dynamically updating concentrations in cell media, a strategy that would not translate to in vivo application.
It was here we said, "If we truly believe in our hypothesis, which is that the biophysical state of the cell is a more effective lever to control biology, why not just buy a signal generator, an amplifier and a parallel plate capacitor to create a uniform electromagnetic field over our cells to do programmed frequency screening?" So- that's what we did.
There were obvious limitations in this simple set-up but the initial results we acquired perturbing cells and observing changes in calcium signaling, membrane potential and water structure led us to the conclusion I mentioned at the onset of this post. Bye-bye molecules, hello medbed.
You’ll notice this is my first co-authored blog on this platform. @anabology is a great follow on X, but he’s an even better co-founder.
We are actively seeking adaptable engineers aligned with our mission of advancing human longevity to develop hardware at the intersection of acoustics, magnetism, and RF. Interested folks can learn more by reading our whitepaper.
Realize the inconceivable.
-Benjamin Anderson
Nostr: ben@buildtall.com
Here you might be thinking that probability-based screening assumes an exhaustive random search, ignoring targeted drug design. Still, conservatively, even if targeted design improves the hit rate by 10,000-fold through structure-based optimization and AI-guided libraries, we'd need to screen around 18.7 million compounds, at a total cost exceeding $9.35 trillion for assays alone, before human validation.
 | A guest post by
|
Nice post!
Regarding the “magic molecule” probability, what you calculated here is the tail of a model you assumed, which treats lifespan effects as if all possible compounds follow a single normal distribution, and the DrugAge compounds are random samples from it (or a biased sample, doesnt change the core issue). But I doubt that's the case, that compounds follow such a distribution... It's a rather strong (and pessimistic) assumption.
That said, I think there’s a good reason to question if a single molecule could cause such a big effect, because it would have to modulate so many processes at once. Moreover, if such a molecule existed and were accessible, evolution should have discovered and kept it. Small lifespan tweaks can hide behind trade-offs, but a 50% boost is extremely hard to beat in terms of fitness, unless some other force, e.g. population-level selection, makes lifespan extension undesiarable.