cryonics-posts post 2 draft-post

The real bottleneck in brain preservation is not the cryoprotectant

Why bloodflow, capillaries, and the choice of imaging modality are load-bearing for evaluating brain cryopreservation — and why "looks fine on CT" is an epistemic trap.

Epistemic status: I met with Nectome (Aurelia, Charlie, Jasmine) in Portland yesterday and spent the evening trying to check what they told me against the literature. I think the broad picture is right, but I want to flag the specific numbers I could and couldn't verify. Confidence tiers: C1 = primary source / peer-reviewed, C2 = credible secondary, C3 = my inference, C4 = anecdote, C5 = could not verify.

The 14-minute puzzle

The Nectome protocol for physician-assisted-death (MAiD) preservation has a specific number at the center of it: 14 minutes. If cryoprotectant washout perfusion is initiated within about 14 minutes of cardiac arrest, Nectome's 2026 bioRxiv preprint shows that pig brains come through with intact membranes, visible mitochondria, and "perfectly preserved synapses" at volume-EM scale. At about 18 minutes: "evident cellular damage." This is less a smooth gradient and more a cliff. [C1]

Why would there be a cliff, and why at 14 minutes specifically?

The intuitive model most people carry around is something like "tissue slowly dies, we perfuse cryoprotectant slowly, race against the clock, no sharp transitions." That model is wrong, and the reason is where the death happens: in the capillaries, before it happens anywhere else that matters.

This is a post about why I think the perfusability of cerebral capillaries, not the chemistry of cryoprotectants, is the real limiting factor in brain preservation quality — and about why the modality you use to evaluate your preserved brain determines whether you are even looking at the thing you care about.

Poiseuille and the tyranny of small pipes

Start with the physics. Laminar flow through a pipe is governed by Hagen-Poiseuille:

Q = (π · ΔP · r⁴) / (8 · η · L)

Flow ∝ radius to the fourth power. Halving the radius drops flow to one-sixteenth. [C1]

Now add biology. Human brain capillaries have inner lumen diameters in the 2–5 μm range — the outer wall puts the whole tube at about 7–9 μm (Chakraborty et al. 2023). Red blood cells are 7–8 μm across. In much of the cerebral microvasculature, the RBC is larger than the capillary it's traveling through. This is a feature, not a bug — RBCs deform into parachute or bullet shapes, which maximizes surface contact for oxygen exchange and actually uses oxygen-sensitive membrane mechanics to autoregulate flow (Wei et al., Neuron 2016). [C1]

But the geometry has no safety margin. Shrink a capillary lumen by 2 μm and nothing gets through. Not the RBC, not a stuck neutrophil (neutrophils are 12–15 μm, even less forgiving than RBCs), not a 9-molar viscous soup of DMSO, formamide, and ethylene glycol you're trying to push in as cryoprotectant.

Aurelia said something like "in big vessels the pressure is fine; in the capillaries a tiny swelling means no flow." I'd frame it more starkly: cerebral capillaries exist at the edge of passability in health. They have no tolerance for even partial narrowing.

The human brain has something like 400 miles of these vessels — ~640 km of capillary network inside a 1.4 kg organ, every neuron sitting within ~40 μm of a capillary (Cipolla, "The Cerebral Circulation"). That density is the reason oxygen reaches everywhere in health. It is also the reason that when things go wrong, they go wrong uniformly at the microscale and undetectably at the macroscale. [C2]

No-reflow: why waiting doesn't work

In 1968, Ames and colleagues published Cerebral ischemia. II. The no-reflow phenomenon — one of those experiments that gets cited forever because it answered a clean question with a clean method (Ames et al. 1968). They inflated a 350 mmHg cuff around rabbit necks to block both carotids and the basilar artery, held it for 5–15 minutes, then tried to reperfuse with a carbon-black stain and asked: how much of the brain took the stain?

The answer: at 5 minutes, patches of tissue don't re-perfuse. At 15 minutes, almost the entire cerebral hemisphere fails to re-perfuse — the blood stain just stops somewhere upstream. You can't argue the flow back in with higher pressure, because at that point the capillaries have closed at their own level. [C1]

The modern version of Ames is more mechanistically detailed and less dramatic about any one mechanism. Four things happen in parallel (Zhang et al. 2024; Kloner 2018):

  1. Pericytes contract and then die in rigor. Pericytes are the contractile cells wrapped around capillaries. Hall et al. 2014 (Nature) showed that under ischemia pericytes actively clamp, and then — denied ATP to release — they die in the constricted state. Irreversibly. [C1]
  2. Astrocyte end-feet and endothelial cells swell from cytotoxic edema (ion-pump failure → intracellular Na⁺ and water). They're pressed against the capillary from outside. Capillary diameter drops from both sides.
  3. Neutrophils plug capillaries. El Amki et al. 2020 (Cell Reports) showed 20–30% of capillaries in mouse ischemic core are stalled by neutrophils specifically, causing tissue reperfusion to cap at ~60% of baseline even after successful clot removal. [C1]
  4. Red blood cells get trapped in columns where the capillary has constricted, forming micro-plugs.

Aurelia told me capillaries "collapse by about 90% in volume" during ischemia. The actual literature is subtler: individual capillaries at pericyte sites show ~37% diameter reduction (≈60% volume loss at that location), 20–40% of capillaries stop flowing entirely, and the net effect is that most of the brain becomes unreachable through the normal arterial tree within 15 minutes of normothermic arrest. So "90% in volume" isn't literally what the numbers say, but the operational conclusion — you cannot reach the brain through its own vasculature after ~15 minutes of warm ischemia — is exactly correct. [C2 on conclusion, C3 on the specific 90% figure not being quite right but close enough in spirit]

This also explains the cliffness of the 14-minute window. This isn't tissue dying on an exponential; it's a network undergoing a percolation-like failure. At t=13 the network is mostly passable; at t=18, you've lost continuous reach to large fractions of parenchyma.

The skull is part of the problem

The rigid calvaria adds a second, nastier failure mode. The Monro-Kellie doctrine says the skull's total contents — brain, CSF, blood — sum to a constant volume. Normal brain is 77–80% water. A 1–2% increase in brain water raises intracranial pressure hugely once the venous and CSF buffers are exhausted. Once ICP rises, cerebral perfusion pressure (= MAP − ICP) collapses, which means you have to push harder on the inlet to drive flow, which means the already-compromised capillaries get squeezed or burst.

Aurelia noted in passing that you can safely shrink a living brain by about 10% — mannitol and hypertonic saline do this routinely in neurosurgery — and that a burr hole is well-tolerated. Both are true. The clinical numbers say brain volume can be reduced several percent with osmotherapy, and 10% is the aggressive upper bound you'd only see stacking agents. Decompressive craniectomy of 12–15 cm is standard for refractory ICP elevation and reduces mortality substantially in malignant stroke and TBI (Frontiers 2019; Sci Rep 2024). [C1]

The right framing is: the brain itself is not fragile; the skull is what makes pressure deadly. And since cytotoxic edema starts forming within minutes of ischemia, the vault constraint is doing damage during exactly the window you need to perfuse.

This is, I think, an underappreciated component of ASC's architectural advantage over vitrification-only approaches. Aldehyde fixation in the very first step of ASC cross-links proteins in the vessel walls themselves within seconds. A glutaraldehyde-stabilized capillary resists further swelling. It also resists barotrauma when you later ramp the pressure to push a viscous 9-molar cryoprotectant through. This isn't a mechanism I've seen people foreground, but Andrew McKenzie's 2024 review of the field (PMC11416988) names it directly:

"Aldehyde fixation... likely involves... (b) stabilizing blood vessels to improve cryoprotectant perfusion." [C1]

That is the sentence I wish someone had put in bold in the cryopreservation literature a decade ago.

Cryoprotectants do not solve the perfusion problem

When the question is "how do we protect already-loaded tissue from ice," modern cryoprotectants — M22, VM-1 — are remarkable pieces of engineering. M22 is a ~9.3 molar cocktail of DMSO, formamide, ethylene glycol, N-methylformamide, 3-methoxy-1,2-propanediol, polymers, and two synthetic antifreeze-protein analogs (biostasis.com on M22). It can vitrify whole organs. [C1]

What M22 cannot do is push itself into a collapsed capillary. Cryoprotectant physics assumes the vessels are open. Take that assumption away — which is exactly what happens when your patient has been dead for 18 minutes under normothermia — and the cleverest CPA in the world can't reach the inner few millimeters of a cortical gyrus. McKenzie et al.'s review frames this honestly:

"Postmortem perfusion faces the critical 'no-reflow phenomenon.'... How rapidly cerebral perfusion degrades in the postmortem period... is an open question." [C1]

ASC's answer is to run fixation as the first perfusion, not the last: get glutaraldehyde into still-open capillaries, stabilize everything, then take your time with the cryoprotectant ramp. That's the 2015 McIntyre & Fahy paper that won the Brain Preservation Foundation small-mammal prize. That's what Nectome is extending to a pig brain in 2026, with perfusion starting within 14 minutes of a MAiD-simulated arrest.

The Sparks Brain Preservation counter-position is: if the CPA ramp is where most of the expense comes from, and fixation is cheap and already enough to preserve structure (as a century of neuroanatomy has shown), just do the fixation and store at −20 °C. There is truth in this. There is also a missing head-to-head: no one has published a direct EM-scale comparison of a brain preserved by Sparks-style low-cost field fixation against a brain preserved by Nectome-style careful perfusion + vitrification, evaluated by the same independent lab, on matched timing. The cost difference is real; the quality difference is genuinely open. [C5 on the head-to-head, C1 on the cost difference]

I'll give Aurelia's version of this disagreement because I thought it was good: Sparks thinks cost is the bottleneck, so you should get preservation to as many people as possible as cheaply as possible. Nectome thinks quality is the bottleneck, because "it doesn't matter if the information is gone, and we have no way of knowing." Both positions are internally consistent. The empirical question — how far below BPF-grade preservation is Sparks-grade, on matched ischemia times? — is not resolved.

The scan-modality trap

This is where epistemics get dangerous.

Aurelia's way of putting it: CT can look fine, optical microscopy can look fine, and the electron microscope can still show you catastrophic ultrastructural damage. She's right, and the physics say it has to be this way.

Here's the resolution ladder (Nature Rev Neurosci 2025 on connectomics; Nature Biotech Lichtman 2023):

Synaptic clefts are 20 nm. Postsynaptic densities are 30 nm. Fine axonal branches can be <100 nm in diameter. To trace the connectome, you need pixel sizes on the order of the thinnest axon with some oversampling — which is why the BPF Large Mammal Prize specifies 5 nm EM pixels and "every synapse traceable" (BPF prize rules). [C1]

CT of a vitrified brain is genuinely useful for one specific thing: did the brain vitrify, or are there ice regions? Ice is density-different from vitrified CPA-water glass. Alcor's CT protocol maps this straightforwardly as orange (vitrified) vs red (extensive freezing) (Alcor CT page). [C1]

CT is not useful for answering "is this brain's connectome preserved?" That question is 10,000–100,000× beyond CT's spatial sensitivity.

So: imagine you're Alcor or Tomorrow Bio, and the only quality data you routinely generate is CT. Your brains look orange across the board. By the evaluation instrument you have, everything is fine. And now Tomorrow Bio's own 2026 roadmap says:

"Since 2024, Tomorrow Bio cannot improve much anymore based on CT scan alone, as close to 100% of the brain shows sufficient CPA concentration for vitrification."

And consequently, they're starting — in 2026 — to take 2–3 brain microsamples from consenting patients for electron microscopy (Tomorrow Bio 2026 roadmap; founder letter). Their own words: "Good CT scans are a necessary condition, but not a sufficient one." [C1]

That is, in 2026, one of the two largest for-profit cryonics operators is publicly acknowledging that their main quality-evaluation tool is orthogonal to the quantity they care about.

Aurelia's complaint about Tomorrow Bio — paraphrased — was that they aren't running "rat left out for X minutes then preserved then EM'd" experiments. That is, they aren't running the model-organism audits that would calibrate their field protocol against the EM-scale truth. I couldn't find any published evidence that they are. Nor could I find any other vitrification-only provider running the deliberate-ischemia-then-EM paradigm on rats. (If it's happening internally anywhere, it isn't public. [C5])

The only operation publicly running this experiment at useful scale is Nectome, and the 2026 preprint is precisely that experiment in pig. The independent third-party evaluation — Andrew Critch walking samples to Berkeley's EM core, picked with a quantum RNG, where stress-test samples held at 60 °C for 12 hours showed "near-identical" preservation — is unusually rigorous (Nectome LessWrong post). [C1]

The epistemic trap is specific: a provider can do everything right according to its instruments and still be producing nothing of value. The modality is load-bearing. A provider that doesn't invest in EM audits has no principled basis for its quality claims; it only has "didn't visibly fail on CT," which is not the same thing.

This is not an anti-vitrification point — ASC does a cryoprotectant ramp too. It's a point about audit philosophy. Nectome audits with EM at pig scale. Alcor and Tomorrow Bio (as of 2026) audit with CT at whole-head scale. The philosophical differences between ASC and vitrification-only aside, that difference in audit regime matters — maybe more than the difference in protocol.

The 19-factors point and epistemic humility

Aurelia said "something like 19 factors affect ischemia, and nobody controls all of them." I spent an hour trying to find the 19-factor list and I don't think it exists as such. What does exist is a literature describing 30–40 partially independent mechanisms by which an ischemic brain kills itself: ATP depletion, Na⁺/K⁺-ATPase failure, glutamate excitotoxicity, NMDA/AMPA-receptor overactivation, calcium overload, ROS production, lipid peroxidation, mitochondrial permeability transition, ferroptosis, necroptosis, PANoptosis, calpain cleavage of cytoskeleton, lysosomal permeabilization, BBB tight-junction degradation, cytotoxic edema, vasogenic edema, pericyte constriction, neutrophil plugging, microthrombi, spreading depolarizations... (Qin et al. 2022, Nature STTT; Siesjö 1988). [C1]

So the spirit of the "19 factors" claim is correct — the exact number is decorative.

The therapeutic record against this cascade is chastening. Decades of trials targeting single mechanisms — NMDA antagonists, free-radical scavengers, Ca²⁺ blockers, magnesium — have mostly failed. The only broad-spectrum intervention that works is hypothermia, because it slows all of it at once.

ASC is an end-run around the cascade. You don't need to block excitotoxicity, ferroptosis, and calpain activation separately; you just need to lock everything in place with glutaraldehyde cross-links before it has time to self-destruct. Fixation doesn't care why tissue was about to die. It turns every mechanism off at once because it turns every protein into a cross-linked polymer that can't be cleaved, denatured, or translocated.

Which means, once again, the problem reduces to: can you get the fixative into the capillaries fast enough? That's the leverage point. Everything else is subsidiary.

Why skin cooling can't save you

A small but important coda. One intuition people have is: can't you just put ice packs on the patient's head in the minutes after arrest to slow the damage?

The physics says no. Brain thermal conductivity is about 0.5 W/m·K; fat is about 0.23 W/m·K; the skull is around 0.32. The time constant for heat to diffuse 5 cm (scalp to midbrain) is about 5 hours by pure conduction. In practice, the brain's 20% cardiac output is a heat source that dominates everything until circulation stops — and once circulation stops, the cooling pathway through blood vanishes exactly when you need it. Transcranial cooling studies show you can drop deep brain 1–2 °C, not the 10 °C that would really help (PMC2094117; PubMed 9754976). [C1]

The only fast cooling is through the circulation — pump cold fluid in via the carotids. Which is to say: cooling and perfusion are the same procedure. "Cool the brain before starting the main procedure" isn't really a distinct option. As Aurelia put it, you may as well start the main procedure.

Q10 for brain metabolism is about 2.3 in the 28–38 °C range (steeper, up to ~12, near body temp — the first 5 °C of cooling buys disproportionately much). So extracorporeal cooling to deep hypothermia can extend the tolerable warm-ischemia window by 5–10×. This is the logic behind DHCA for aortic surgery. For cryonics, it's mostly academic; for MAiD-compatible preservation, it could be the single biggest regulatory lever the field has, if pre-arrest partial cooling were ever permitted. [C2]

Cost vs quality: Sparks vs Nectome

The field has a genuine philosophical fork.

Sparks Brain Preservation: aldehyde fixation is a century-old gold standard for neural structure. Fluid preservation at −20 °C is robust and indefinitely cheap. Preserve as many people as possible for as little as possible. If 90% of the information is there, that's much better than 0%, which is the outcome for everyone who can't afford cryonics.

Nectome: we don't know what fraction of information is there, because most of the ways of checking cost more than the preservation itself. Until you've audited it with EM, "the brain looks fine" is indistinguishable from "you have no data." Therefore invest in quality: slower procedure, higher fidelity, independent third-party EM, and treat the cost ceiling as a downstream problem to solve after quality is established.

Neither position is obviously wrong. Sparks's argument is actually quite strong once you notice that any structural preservation is better than the default of biological decomposition, and that EM quality above some unknown threshold might not matter (nobody has revived anyone; we don't know the threshold). Nectome's argument is stronger once you notice that nobody can tell you where the threshold is without the very EM audits Nectome is doing.

The piece of evidence that would move me most is the one that's missing: a matched-timing head-to-head EM comparison between Sparks-grade and Nectome-grade preservation, on rats, evaluated by an independent lab. Someone should fund this. If it shows that Sparks-grade preservation is already connectome-traceable, the cost argument wins. If it shows that Sparks-grade preservation is ultrastructurally compromised in ways Nectome-grade isn't, the quality argument wins. Right now, neither side has this data in public. [C5]

What I think I learned

Summarizing what I update on, in rough order:

  1. The 14-minute window is physically meaningful and roughly right. It's set by the timescale of capillary closure from no-reflow, cytotoxic edema, and pericyte contraction — which collectively render most of the brain unreachable through its own arterial tree by ~15 minutes of normothermic arrest.

  2. The r⁴ Poiseuille point isn't an analogy. Cerebral capillaries are at the RBC-deformation threshold in health. Modest narrowing → nonlinear flow collapse → cryoprotectant can't reach tissue.

  3. ASC's key innovation is not the cryoprotectant; it's the order of operations. Fixing first with glutaraldehyde stabilizes both cells and vessels before the high-osmolarity CPA ramp. This is the reason ASC won the BPF Large Mammal Prize and vitrification-only has not.

  4. The Monro-Kellie vault is a big deal during preservation and osmotic brain shrinkage (10%-ish, stacked agents) is a real tool. A burr hole at the right moment might matter.

  5. Scan modality is load-bearing for quality claims. CT resolution is 10,000–100,000× too coarse to see the thing you care about. A provider that isn't investing in EM audits has no principled basis for quality claims — just "didn't visibly fail on the cheap test."

  6. Tomorrow Bio is publicly acknowledging, in 2026, that CT-only QC has saturated. This is a meaningful field update; Alcor has not publicly made the same update.

  7. No published vitrification-only provider has run the "rat-left-out-then-preserved-then-EM'd" experiment. This is the single most valuable experiment the field could run. It's relatively cheap.

  8. The Sparks vs Nectome disagreement is not yet empirically resolvable. The head-to-head EM comparison on matched samples doesn't exist. This should bother us.

  9. ~19 (or 30, or 40) things go wrong in parallel during ischemia, and clinical neuroprotection targeting each one individually has a poor record. ASC's "freeze all chemistry at once with cross-linking" is a systems-level end-run, not a mechanistic intervention, and that's probably why it works when mechanistic interventions have failed.

  10. Cooling the skin doesn't meaningfully cool the brain. This is hard thermal physics and I wish more popular-science coverage of cryonics acknowledged it. The only fast cooling is through the circulation, which is the main procedure.

The through-line: the bottleneck in brain cryopreservation quality is not the chemistry of cryoprotectants. It is getting fluids into capillaries before the capillaries close. Every other technical choice in the field — protocol design, cooling strategy, audit regime — is a downstream consequence of this single physical constraint. Once you see it this way, a lot of the apparent fragmentation in the field resolves: the Sparks/Nectome debate is about what you optimize given this constraint; the Alcor/Tomorrow Bio/Nectome differences in audit regime are about how honestly you're willing to measure yourself against this constraint. And the capillaries themselves, a 400-mile network of 5-μm pipes humming inside your head right now, are doing a quiet, elegant job that has no safety margin at all.

That's worth preserving carefully.

Unresolved questions I'd like to chase

ai gen