Brain extraction damage
In updating the original post, I have added a new list specifically dedicated to the inferential loss introduced by the surgical removal of the brain. This category was not present in the previous version and deserves its own treatment. The idea emerged partly from the discussion on
robotic surgery — which highlighted how future surgical precision may reduce extraction related damage — and partly from
your observation that fully successful perfusion is expected to be an exception rather than the rule. When perfusion is incomplete, small pockets of unfixed water may remain in the cortex even after hours of formalin‑based perfusion. These pockets would freeze at −20 °C and cause severe local damage; in some cases they may also reflect tiny regions that never received full perfusion, producing a very small amount of residual ischemia during the perfusion phase. Removing the brain allows full immersion and eliminates these “shadow regions”, which is why extraction becomes
necessary in most real‑world cases.
Although this section focuses strictly on the mechanical inferential loss introduced by extraction itself, it is worth noting that future improvements in perfusion technology — such as multi site robotic perfusion (i.e., perfusion initiated from multiple anatomical access points), ventricular perfusion, subarachnoid perfusion, or epidural/subdural cooling — could reduce the frequency of incomplete perfusion. These approaches are technically demanding and potentially risky if performed poorly. The effects of these perfusion related techniques are conceptually distinct from extraction damage, and their small potential benefits (likely well below 0.1% in LTM inferential terms, though possibly up to twice as large under Alcor/CI protocols) are already accounted for in the warm ischemia and on perfusion cold ischemia lists; they are therefore not quantified again in this section.
Rather than modeling visible tissue destruction, the estimates reflect how different extraction techniques alter microdomain stability, synaptic registration, and surface‑level continuity in ways that slightly reduce the theoretical reconstructability of pre‑mortem states.
Cracking [and ice]
Contrarily to the brain extraction damage, which is strictly superficial and limited to cortical micro‑regions, cracking and ice damage disrupt deep volumetric continuity and therefore operate on a completely different inferential scale. Fortunately, in SBP protocols the chemically stabilized brain does not form the brittle, cooling-induced glassy matrix produced by high-concentration vitrification solutions in Alcor/CI protocols; the fixation cross-linking itself prevents the formation of a fragile amorphous solid, so only accidental mechanical shocks during storage remain a concern.
You suggested that cracking might be “
essentially zero loss”, but the justification you added (“
in spite of my guess about pulverization”) only explains a reduction in confidence, not the conclusion itself. Likewise, the shift to “
the real concern is ice” does not justify the assumption that cracking is negligible.
My concern was not ice in SBP protocols because
you told me ice is 0% there. But for Alcor/CI, ice is a real factor, so I prepared a structured list.
Alcor/CI long term-memory inferability-loss from ice damage
(post-2009 cases; CNV= Cryoprotectant Net Volume, not strictly equal to “non-ice” because it can include under-perfused voxels, sub-vitrification, or local CPA gradients)
- Top 22% Alcor cases (CNV 96.5% ± 2.0%): 4–8% (ice ~0.1–0.5%, “essentially ice-free” (Fahy & Wowk, 2021), functional impact remains amplified by microdiscontinuities; even minimal fractures interrupt synaptic chains and neuroendocrine nodes in CA1/hypothalamus, producing losses greater than volumetric estimates; local mechanical propagation → 0.5–1.0%; intrinsic osmotic damage from freezing: intracellular dehydration and extracellular compaction during glass transition → 0.2–0.4%).
- Top 50% Alcor head-only (CNV 85% ± 3%): 7–12% (ice ~1–2% confined to ventricles and subarachnoid spaces; CNV higher and more stable than whole body due to smaller thermal mass and targeted perfusion; microfractures in CA1/hypothalamus with amplified impact on LTM → 3–5%; local propagation → 1–2%; intrinsic osmotic damage: synaptic compression and extracellular matrix alteration from water-solute redistribution → 0.5–1%).
- Top 50% Alcor whole-body (CNV 78% ± 5%): 11–18% (ice ~2–4% with peripheral macrocrystals and diffuse microfractures; CNV lower and more variable than neuro cases, with systemic cooling stress and less targeted perfusion; CA1/hypothalamus interruptions aggravated by volumetric gradients → 6–9%; extended propagation/recrystallization → 2–3%; intrinsic osmotic damage: solute gradients and cellular shrinkage during deep cooling → 0.5–1%).
- Top 50% Cryonics Institute (CNV 72% ± 6%): 14–22% (ice ~3–5% with less uniform distribution and broader nucleation; perfusion/cooling protocols less optimized than Alcor; CA1/hypothalamus discontinuities more severe, amplifying functional loss → 7–10%; propagation/recrystallization more extensive → 2–3%; intrinsic osmotic damage: stronger shrinkage and solute concentration due to less controlled CPA/thermal profiles → 1–1.5%).
NOTE: Neocortical damage is implicitly included via volumetric ice and osmotic terms but is not anatomically isolated due to distributed and partially redundant encoding. Hippocampal CA1, entorhinal cortex, and hypothalamus act as high-centrality bottlenecks and are disproportionately vulnerable.
I’m not sure whether these ice‑related damage percentages I listed are as “impressive” as this
statement of yours: “
As ice expands, it crushes and smears the tissue with a force of about 75,000 psi. Any physical process that stirs or smears molecules causes a kind of damage which would not allow future inference of the original structure. You just can't unmix something. There's not enough information.”
Since your description implicitly treats ice damage as fully non-invertible, I will temporarily adopt a high safety-margin as well: I have generally used a global safety-margin of ~40–50% for my new percentages, but for cracking I will use ~60–70%. Note that, unlike cracking, ice does not admit any “planar” or nearly lossless scenario: volumetric expansion and recrystallization cannot produce smooth or re‑alignable interfaces. There is no benign hypothesis for ice: only the brittle, chaotic, non‑invertible regime exists. Cracking instead requires a different treatment because two physical regimes remain plausible: the “planar” vs “brittle” hypothesis. To explain this point, I prefer to “
expand on cracking” as well.
Cracking introduces a different inferential problem:
loss of continuity in distributed engram pathways. LTM is not a static mosaic but a distributed code across hippocampus, prefrontal cortex, amygdala, and cortical ensembles, where recall depends on coordinated integration across regions (
Tonegawa et al., 2015). You correctly noted:
“Neurons have long delicate axons and dendrites … a single neuron can span the entire brain … matching two ends could be ambiguous … A molecular fingerprint could allow matching up two ends in a damaged area.”
This is true — but only solves
fiber identity, not
fiber destination. If the fracture occurs at the point of arrival, the system loses information about
where that process was supposed to terminate. More generally, if the fracture disrupts a considerable volume around an axon or dendrite, even if you can match their two ends, you still do
not recover:
- which synapse it was forming
- which spine head was the target
- which micro-cluster or micro-column it belonged to
- which high-centrality node (CA1, hypothalamus) it linked into
- which vesicle-distribution pattern characterized the active zone
- which perisynaptic glia were part of the micro-circuit
These features are long-term memory. Matching the two ends of a fiber does not reconstruct the topology of its synaptic embedding — and this can be even more relevant than losing an entire less-central neuron.
For the purpose of inferability analysis, any connectomics and network-based models of memory should probably decompose into three interacting components:
T = topological disruption — interruption of engram‑spanning pathways and high‑centrality nodes
I = interface uncertainty — irregular, micro‑branched fracture surfaces and small uncertain volumes
S = synaptic embedding ambiguity — uncertainty about which spine, micro‑cluster, micro‑column, vesicle pattern, or perisynaptic glia constituted the original synaptic target.
Cracking increases all three terms simultaneously: Inferential-loss = T × I × S. For the updated percentages, the AI modeled these components in a way that can be approximated on a rough 1–5 scale:
Localized microcracking → T=2, I=2, S=3
Diffuse microcracking → T=3, I=3, S=4
Macrocracking → T=4, I=4, S=5
We still need to clarify how to treat the previously mentioned “two physical regimes” of cracking. Ultramicrotome sectioning produces smooth, controlled surfaces that preserve ultrastructural continuity with high confidence. By contrast, brittle macro- or microcracking in vitrified tissue is not equivalent to controlled sectioning. While direct neuroscientific studies have not documented pulverization or debris loss, analogies with glassy materials indicate that fracture propagation typically generates irregular, non-planar interfaces through well-known mechanisms such as crack-front instability and microbranching. In glassy solids, the absence of visible pulverization does not imply planar or symmetric fracture surfaces. Pulverization is a macroscopic manifestation of energy dissipation, but the same energy can be dissipated during crack propagation through modes that remain below the resolution of naked eye inspection:
- sub-micron crack bifurcations
- surface roughening
- microbranching
- sub-micron chipping
- nanoscale debris
- shear-induced amorphization
- sublimation-like transitions
Vitrified tissue, which mechanically behaves as a brittle glass, is therefore expected to fracture through unstable crack-propagation modes that produce rough, non-alignable interfaces even when no loose fragments are macroscopically detectable. Consequently, observing a clean macroscopic split cannot be taken as evidence for planar cracking at synaptic scales. Each uncertain fragment or rough micro-gap can generate non-zero uncertainty at synaptic scales. Operationally, only the macroscopic geometry of the crack is visible during handling; the micro-scale features that determine inferability remain concealed within the macrocracking event itself.
Within the mentioned 60–70% safety-margin band, I keep the estimates anchored to the
inferability perspective that even small cracks may have a disproportionate impact. Since the planar scenario has been proposed as a potentially low-loss regime, I model cracking under two mutually exclusive hypotheses:
- Planar hypothesis. Cracks behave almost like ultramicrotome cuts (nearly lossless at synaptic scales). This hypothesis remains plausible because macroscopic observations in vitrified human brains do not contradict it, while current cryobiological work on cracking — rather than characterizing microscale interface structure — remains primarily focused on minimizing thermal-stress risks for a cracking essentially inevitable in whole-brain vitrification (at least “presently”, in Wowk’s 2011 words, which also specify that cracking “probably does not compromise” LTM inference, likely justifying this “probably” via the planarity hypothesized here).
- Brittle hypothesis. Cracks behave like brittle fractures in glassy materials (rough, micro‑branched, with small uncertain volumes). This hypothesis is grounded in the broader physics of glassy solids, where unstable fracture modes and non‑planar interfaces are common (Lawn, 1993). Consequently, a brittle-type interface remains compatible with the same inevitability assumed in the planar hypothesis, also under macroscopic cooling conditions that appear well controlled — a point that is essential for interpreting the updated cracking inferability‑loss percentages in the framework.
This also connects to the PLOS1 survey, where 70.5% of participants agreed that LTM is highly dependent on structure and synaptic strengths, yet the median probability that LTM could be extracted from a static snapshot was only ~40%. In a worst‑case reading, this leaves ~60% residual uncertainty — which is exactly the range of safety-margin I am adopting here.
Since the goal is not to estimate probabilities but to produce safe estimates under the worst plausible hypothesis, I apply a conservative logarithmic reduction, subtracting ~10% to one of my previous damage estimates, corresponding to a non-informative 50% prior — essentially a
coin-flip assumption between the planar and brittle scenarios. If future evidence were to show that the planar hypothesis holds with high probability — for example, on the order of 90% — then the brittle-oriented estimates would be reduced accordingly. The exact reduction is not fixed here, since it depends on the strength and structure of the empirical evidence, but the framework is designed to update conservatively as the probability of the planar scenario increases. The resulting percentages therefore reflect a high safety-margin, worst‑case‑compatible estimate of maximum inferential loss.
Osmotic shrinking/swelling
This time I agree with this reduction of confidence:
“
… ramped through cryoprotectant, which itself could be damaging. It's known that osmotic pressure can cause damage, …” (
Oct 03, 2023)
“
I wouldn't worry at all about cryoprotectant osmotic or toxic damage. I think that damage is essentially zero.” (
Dec 10, 2025)
Osmotic shrinking and swelling occur whenever a fixative or cryoprotective solution enters the tissue with an osmolarity or permeability different from the intracellular environment. This includes perfusion with formalin (10% NBF ≈ 4% FA), glutaraldehyde (2–2.5% GA), low-concentration DMSO solutions such as those used in SBP protocols, and high-concentration vitrification mixtures such as M22 (Alcor) and VM-1 (CI), as well as earlier glycerol-based mixtures and modern ethylene-glycol–based mixtures.
In all these cases, neurites, spines, and microdomains undergo transient deformation during loading and equilibration. However, deformation alone does not imply inferential loss. As long as the process remains continuous and invertible, the relative geometry of synapses, active zones, vesicle pools, and perisynaptic glia is preserved up to a coordinate transform. Shrinking and swelling changes shape, not structure. For this reason, the osmotic contribution to LTM inferability-loss should already be ~0% at the resolution of this framework.
A hypothetical regime does exist in which osmotic stress exceeds the elastic limit of membranes and cytoskeleton, producing tearing, fusion, cavitation, or irreversible crossings of thin processes. In that regime, osmotic damage could in principle be worse than cracking, because it would not merely introduce interfaces but locally destroy or fuse fine structure.
The point at which deformation is no longer continuous and reversible includes:
- membrane rupture
- spinal detachment
- cytoskeletal tearing
- cavitation
- massive blebbing
- irreversible surface fusions
That is, when you're no longer "stretching" a viscoelastic tissue, but tearing, crushing, gluing, and breaking it.
However, such a regime is incompatible with the ultrastructural appearance of formalin-fixed, glutaraldehyde-fixed, or CPA-treated brain tissue: we do not see widespread membrane rupture, spine fusion, or non-reversible crossings.
Real-world perfusion protocols operate far below this threshold.
Even so, within the continuous regime, a non-equilibrium, non-uniform osmotic state could in principle make the mapping from pre-stress to fixed morphology non-injective: multiple pre-stress configurations may converge to the same final geometry. But if such underdetermination existed at the 0.1–1% level, it would already be visible in EM as inconsistent or collapsed microstructure. The absence of such signatures implies that any inferential ambiguity must be several orders of magnitude smaller than spine destruction or ice/cracking damage, and therefore below the noise floor of this model.
Although the absolute osmotic contribution to inferability-loss is ~0%, different perfusion and CPA protocols impose different magnitudes of osmotic stress. In the continuous, sub-elastic regime, any infinitesimal inferential ambiguity
scales proportionally with the amplitude and duration of these gradients. For this reason, it is convenient to express osmotic effects as simple multiples of a reference infinitesimal ε rather than as explicit percentages.
Let ε denote the infinitesimal osmotic inferability-loss associated with standard formalin perfusion. Other protocols can be placed on a relative scale:
- Formalin perfusion (10% NBF): ε
- Glutaraldehyde perfusion (2–2.5%): ~1.5 × ε
- SBP Low-DMSO solutions: ~2 × ε
- Alcor M22: ~2–3 × ε
- CI VM‑1: ~2–3 × ε
- Glycerol‑based mixtures: ~2–3 × ε
- Ethylene‑glycol–based mixtures: ~1.5–2 × ε
These values express relative osmotic stress, not absolute inferential loss. In absolute terms, osmotic shrinking/swelling contributes ~0% to LTM inferability-loss, and any residual ambiguity is dominated by other damage modes such as ice or cracking.
Cryprotectant toxicity
AI estimates that cryoprotectant toxicity contributes 3–7% LTM inferability-loss in Alcor protocols, reflecting persistent chemical interactions with proteins, lipids, and membrane-associated water in unfixed tissue. In contrast, low-DMSO formulations such as VM-1, used by CI, reduce these interactions and are better modeled as 2–5%, but this reduction — driven by their lower DMSO content — also lowers protection against perfusion-related ice formation. Experimental low-DMSO cryoprotectants developed for different purposes, such as the formulations reported in recent Chinese research (
JoVE 2025), may fall in the 1–3% toxicity range for the same reason, but are not suitable for long-term vitrification. A xenon-augmented vitrification cocktail, though affecting directly neither vitrification process nor ice protection, could push the chemically mediated component of inferability-loss toward the ~1% range: xenon is osmotically neutral, chemically inert, and its effects are overwhelmingly reversible, leaving no detectable structural or molecular footprint above the 0.01% noise threshold. Reaching this ~1% corresponds to optimized EG/PG-based vitrification chemistry operating at the minimum concentrations compatible with threshold freezing-point depression and stable vitrification. For SBP protocols, once the tissue is chemically fixed, cryoprotectant toxicity becomes effectively irrelevant for inferential purposes.
A separate consideration concerns ice nucleation. As noted by Wowk:
“However a phenomenon called ice nucleation happens at a high rate near the glass transition temperature, and in some studies doesn’t become undetectable until 20 degrees below it. Ice nucleation — the local reorientation of water molecules into nanoscale ice crystals — doesn’t cause immediate structural damage. However it can make avoiding ice growth and associated structural damage during future rewarming more difficult. The extent and significance of ice nucleation in highly concentrated cryoprotectant solutions is still poorly understood.”
If long-term storage occurs at temperatures well below Tg − 20 °C, nucleation becomes thermodynamically suppressed and remains inert for centuries, and the resulting inferability-loss is expected to remain within the same negligible losses discussed for chemically fixed brains over 200-year horizons at −20 °C, while here the scenario concerns brains vitrified far below −135 °C — in particular those with cells ready for the “revival” as in Alcor/CI protocols. In such a scenario, the biological burden would be limited primarily to the residual ischemic damage accumulated before hypothermia arrested metabolic activity: a form of injury that future medicine might plausibly reverse without requiring full molecular reconstruction.
However, reaching this deep-storage regime requires crossing Tg during cooling, and for a macroscale organ such as the brain this transition produces mechanical stresses that make cracking practically unavoidable with current technologies. For this reason, one cannot “choose” to fight nucleation instead of cracking: the path that eliminates nucleation necessarily passes through the temperature range where cracking dominates. In this sense, the classical cryonics ideal of a true “revival” from long-term vitrification — without ice, without cracking, without molecular reconstruction, and without a revival-incompatible fixation — fractures under its own physical constraints.
