I want a technology which, if built, would give someone I love a fighting chance if they were on their deathbed. Short of curing all known diseases, it’s hard to find technical prompts which could conceivably measure up to this.
What if you had a TARDIS box, a one-way time travel machine that could take you to the future, when you were sick?
On the y axis, years of life expectancy for someone born in a given year. On the x axis, year of birth. Red interval to illustrate the TARDIS thought experiment.
This figure was inspired by the increase in life expectancy in cystic fibrosis patients over time. more info
Wouldn’t it be great if we had a master dial for medical progress? Imagine being able to turn a knob up, and make medicines arrive at any speed. Unfortunately, R&D pace is hard to predict, and we can’t guarantee a cure will arrive in time for patients in need.
What if, instead of speeding up the arrival rate of medicines, we slowed down time for the patient?
We can already pause biological time for human embryos , small organisms such as C. elegans , and rat kidneys .
Graph of cell number and size for biological tissues, with notes on when they were first reversibly cryopreserved historically. See more info for cell number and length calculations.
In part 2 of this essay, I’ll cover engineering principles behind scaling up these methods for human use. So if you’re interested in that, feel free to jump ahead!
But first, I want to describe what feels to me like the underappreciated heart of the problem - the beauty of temperature as a concept, and how it allows us to pause time, abstract over biological complexity, and connect physics and engineering to biology.
In 2025, a human embryo cryopreserved for 30 years was successfully used in IVF. It gave rise to a live birth . It would be plausible to extrapolate, from the below summarized ICMART data, that over a million people have been born who were cryopreserved as embryos previously. Frozen embryo transfers now outnumber fresh embryo transfers (data specific to autologous cryopreservation, with the below caveats about included countries).
| Year | Frozen embryo transfers | Delivery rate | Ref |
|---|---|---|---|
| 2021 | 811,431 | 29.5 | [2] |
| 2020 | 1,053,466 | 32.7% | [3] |
| 2019 | 1,262,727 | 32.6% | [4] |
| 2018 | 1,080,331 | 31.9% | [5] |
| 2017 | 600,375 | 26.1% | [6] |
The frozen embryo transfer #s should be taken as a lower bound globally, as not all countries contributed to ICMART data.
C. elegans worms are today regularly cryopreserved for long-term storage of genetic stocks in research labs , as originally reported by Brenner - from work by Dr. J Sulston - in 1974 .
More recently, rat kidneys were reversibly cryopreserved, rewarmed, and transplanted into a rodent which had no other kidney, which then recovered normal levels of function .
The first reported human pregnancy using a previously cryopreserved embryo was in 1983, reported by Trounson and Mohr. The first live birth - as it turned out, twins - from a cryopreserved embryo was in 1984, reported by Zeilmaker et al. . These embryos were cryopreserved at the 4-16 cell stage . 1n 1985, pregnancy following blastocyst cryopreservation and replacement was reported by Cohen et al .
The first reports of animal cell cryopreservation were in sperm cells. The earliest reported effect of freezing temperatures on sperm cells is, according to a review by Sherman , possibly attributable to Spallanzani in 1776, who (in Sherman’s words) “noted, however, that spermatozoa became motionless during cooling which "reduced them to a state of complete inactivation" from which they revived on rewarming” . Note that I have not been able to locate the relevant quote in the available online translations of Spallanzani’s work (referenced here) . Sherman also cites Mantegazza as, in 1866, reporting human spermatozoa resisted freezing to -15C .
In 1937, Bernstein and Petropavlovski reportedly demonstrated the protective effect of 0.5–3 M glycerol for freezing of bull, ram, stallion, boar, and rabbit spermatozoa to a temperature of -21C . However Polge, Smith and Parkes in 1949 is often cited as a seminal step forward in mammalian cell cryopreservation, as due to chance observation they found glycerol to be an effective cryoprotectant, and published an effective protocol for cryopreservation .
Jump to figure ↑Part 1: Temperature is a dial, a link, and a lever
Most diseases are defined by symptoms, not by a clear molecular cause. Makes sense - biology is too complex (cells have over ~10^11 atoms note, humans have over 10^27 !) to track each atom. So we use heuristics: replace an organ like a modular component, or treat a piece of vasculature like a pipe. Our ability to create medical technology is limited either to things we discovered by accident (such as variolation), or principles we understand (vaccination in the context of germ theory, or manipulating levels of PCSK9 for heart disease).
In contrast, the principle under reversible cryopreservation is completely clear. Stop molecular motion to pause biological time. We know the goal at the molecular level, though the engineering is still unsolved. Moreover, we can import centuries of conceptual work in physics - the hard-won link between our everyday feeling of ‘hot’ and ‘cold’, and the statistical properties of molecular motion. Lowering temperature is associated with lowering metabolic rate , so we then bridge back to the problem statement we posed above - slowing down time for a biological system.
So how does this link between temperature and biological time work? Let’s start with a toy example. In an ideal gas at equilibrium, lowering temperature slows down molecular motion. As temperature decreases, molecular velocity decreases . This looks like slowing down time for the entire system: molecules go to the same places, and meet the same other molecules, just slower.
In an ideal gas at equilibrium, temperature is proportional to the average kinetic energy of molecules in the system . Because the kinetic energy of an object is 1/2mv^2 (where m is the mass of the object and v is the velocity), given all the ideal gas molecules in the model have the same mass, here temperature is proportional to the average of v^2 over all molecules. This means that as you decrease temperature, the average squared velocity of the molecules in the system decreases proportionally.
While this is a compelling model, it should be noted that the concept of temperature comes with some important caveats - it’s well defined at equilibrium , which biological systems are far from. In this context, temperature won’t give us an exact link to average kinetic energy - and correspondingly molecular motion. It’s used ubiquitously, though, in biochemical contexts - for example, in the context of the Arrhenius equation . The Q10 factor comes from a simple manipulation of the Arrhenius equation. \[ \text{rate} \propto e^{-E_a/(k_B T)} \quad\text{(Arrhenius empirical relationship)} \] \[ \begin{aligned} \frac{\text{rate}(T_2)}{\text{rate}(T_1)} &= \frac{e^{-E_a/(k_B T_2)}}{e^{-E_a/(k_B T_1)}} \\ &= e^{-\frac{E_a}{k_B}\left(\frac{1}{T_2}-\frac{1}{T_1}\right)} \\ &= e^{-\frac{E_a}{k_B}\frac{T_1-T_2}{T_1 T_2}} \\ &= e^{-\frac{50~\mathrm{kJ/mol}}{8.3~\mathrm{J/(mol\cdot K)}} \times\frac{-10~\mathrm{K}}{(300~\mathrm{K})^2}} \approx e^{0.7} \approx 2 \end{aligned} \] \[ \begin{aligned} E_a &\approx 50~\mathrm{kJ/mol} && \text{(characteristic activation energy)} \\ T_2 &= T_1 + 10^\circ\mathrm{C} \\ T_1 &\approx 300~\mathrm{K} && \text{(room temperature)} \\ k_B &\approx 1.38\times10^{-23}~\mathrm{J/K} \approx 8.3~\mathrm{J/(mol\cdot K)} \end{aligned} \] adapted from ).
As a sidenote, it’s amusing to read about history of equations linking reaction rate to temperature - it was described as one of the ‘darkest chapters in history’ by Ostwald and it appears that the Arrhenius equation was favored at the time it was chosen for its theoretical simplicity - despite that other equations linking reaction rate and temperature could explain the existing data better .
As a funny note, also from book.bionumbers.org, ants also change speed ~2x for every 10C drop in temperature, around physiological ranges!
Jump to figure ↑In a biological system, the link isn't as perfect. Near physiological temperatures, different reactions slow at different rates. However, in this range, the core principle generally holds. For example, chemists use the Q10 coefficient as a rule of thumb, a heuristic where reaction rates decrease by 2-3x for every 10C temperature drop .
The picture gets more complicated as you get cold enough for phase transitions to occur - in these regimes, we know from the skyrocketing viscosity that molecular motion effectively stops . While this would normally be comforting (stopping molecular motion is effectively like pausing biological time), entering this temperature range raises a new problem we haven’t yet addressed - ice formation!
Part 2: You can turn biology into glass, not ice
Classical nucleation theory, building on Gibbs' thermodynamics, captures a push-pull dynamic. When water molecules form a small ice crystal, they lower the Gibbs free energy proportional to the ice crystal volume but pay a surface-area cost in free energy. This means that very small ice nuclei are thermodynamically unstable: they might form at a high rate, but tend to redissolve. But, past a certain point of ice crystal growth, the volume term dominates, and growth of the crystal is favored. This is illustrated in the equation below - showing Gibbs free energy as a function of ice nucleus radius - which can be solved for the critical radius at which free energy favors growth of the ice crystal.
$\Delta G(r) = 4 \pi r^2 \gamma - \frac{4}{3} \pi r^3 \Delta G_v$
So, as you cool, ice nuclei are constantly blinking into existence, and only occasionally are large enough to be thermodynamically favored to grow!
Ice nuclei have been reported to form in supercooled nanodroplets at rates as high as 10^23 nuclei per cm^3 per s . However, accessing the same ‘no man’s land’ temperature range in bulk-like samples is difficult, and nanodroplet measurements may not reflect microdroplet or bulk behavior. Experiments on micron-sized droplets suggest maximum rates mcloser to 10^16 nucleli per cm^3 per s in this context! Once an ice crystal forms, it can extend at high rates - for example, up to growth rates up to 0.27 m/s were seen in 40-micron supercooled water drops in vacuum after ice nucleation around 234-235K !
The temperature for fastest extension of ice nuclei (once they are formed) lies above that for maximal ice nuclei formation. So, rewarming a biological sample needs to be done more quickly to avoid ice formation than cooling.
Pierre Boutron originally proposed the idea of critical cooling and rewarming rates: the rates required to traverse the ‘danger zone’ without ice formation, given different concentrations of cryoprotective agents .
The ‘danger zone’ for ice formation is characterized as starting below 0C, through to the glass transition temperature for typical cryoprotectant solutions (typically around -130C) at which point viscosity increases to the point where there is essentially no molecular motion to cause ice formation.
Jump to figure ↑At first, it seemed like the beautiful link between temperature and molecular motion was made useless by the way water turns to ice. Biological systems are ~70% water , and normally as you cool, water expands into ice . This causes tissue to tear and kills cells . Luckily, the way that ice forms gives us several places to intervene.
Ice forms in a specific temperature range (the ‘danger zone’), between around 0C and -130C in cryopreservation protocols note, and does so in the form of small, randomly occurring ice nuclei that only sometimes pass the critical size threshold to keep growing . So water doesn’t just immediately freeze solid at a set temperature. If you can traverse this 'danger zone' without ice formation—cooling and rewarming quickly enough or adding chemicals to block ice formation—you can make it to the other side unharmed by ice.
So, what if we just tried to cool biological systems in such a way that they become glass, not ice? This insight (first proposed by Luyet , and demonstrated in mammalian embryos by Fahy and Rall ) creates an interesting engineering challenge. In the method described by Fahy, you load the biological tissue with high concentrations of chemicals (cryoprotective agents, or ‘CPAs’) which block ice formation . But, this creates a new problem - at high concentrations, the chemicals can then cause new types of toxicity over time. So, you want to minimize the concentration required. As shown in a review from Han and Bischof, you can tradeoff CPA levels with cooling and rewarming rate . The faster you cool and rewarm, the less CPA concentration you need.
This creates a clear place to focus engineering efforts in cryo - doing high throughput chemical screens or rational design to find well tolerated cryoprotective agents, and engineering systems to cool and rewarm as quickly as possible within thermal gradient limits. At first, the latter goal might seem daunting. Although vitrification techniques with incredibly high CPA concentrations (40% or more water replaced! ) are regularly used in the clinic today (for example, in cryopreservation of human embryos for IVF) these only involve CPA exposure on the scale of minutes. Embryos are hundreds of cells and similar in size to a human hair . Volume scales as the cube of size; surface area scales as the square. A human organ could require hours to cool convectively , not the minutes an embryo needs. That’s the scaling problem.
Luckily, our bodies evolved a system to circumvent this for nutrient diffusion and heat transport - vasculature. Vasculature evolved to get within a few cell lengths of almost every cell in our body , and methods which take advantage of vasculature may be effective in cooling and rewarming tissue faster - such as perfusive cooling of fluids which keep low viscosity at cold temperatures, or nanoparticle-based rewarming using an alternating magnetic field to transfer energy to nanoparticles which diffuse it homogeneously into tissue.
The protocol possibilities don’t stop there - temperature links you to the vast conceptual vista of physics, often allowing you to check feasibility before experimenting and engineer against clear constraints. You can apply methods from different subfields to manipulate temperature (everything from isochoric cryopreservation through to ultrasound for rewarming ).To make the CPA toxicity problem easier, you can load CPAs at a very cool temperature (for example, 4C), decreasing metabolic rate and correspondingly increasing the time they can be tolerated . You can mimic nature by using or further engineering anti-freeze proteins , which more specifically block ice extension and might decrease the CPA concentration required.
These methods are not hypothetical [LINK TO GLASS NOT ICE] . They’ve led to reversible cryopreservation of small mammalian organs . We (at Until) and others are studying these techniques in the context of reversible cryopreservation for donor organs. Whether a better CPA, or perfusive cooling, or a specific warming method is the optimal way to go will become clear in the coming years, in pig retransplant studies and, if those succeed, later human clinical studies.
These techniques, scaled to human organs, would be transformative for transplant patients and surgeons. Organs expire fast - sometimes in a matter of hours - so when someone donates an organ, the transplant surgeon often charters a private jet to pick it up, sprinting to get there in time. Transplant patients are on call for months - sometimes years - waiting to go into surgery at a moment's notice, living within a two hour radius of a transplant center with a pager, uprooting their lives for the chance at a life-saving organ. Whole organ cryopreservation would save lives, and drastically change quality of life for transplant patients on the waiting list.
This organ work raises a larger question: what is the upper bound on scaling reversible cryopreservation protocols? Scaling vitrification to a whole human body is not provably impossible, but it introduces immense new engineering challenges and runs into unknown questions in neuroscience. You’d need to create CPAs which are not only well tolerated in one organ, but by all organs in the body (or design surgical methods to isolate organs during the procedure). You'd need to solve problems of whole body perfusion. You’d have to develop potentially new assays for neural function to track progress along the way.
And that’s before getting to the core neuroscience question - is it even possible to pause the motion of all of the molecules in a brain and then rewarm it with neural function intact?
At first you might assume this should be obviously impossible, and that even slices of neural tissue - notoriously finicky - should never survive the kind of protocols required for vitrification. But, surprisingly, we now know it's possible to preserve spontaneous and evoked action potentials in acute rodent cerebellar slices , electrical activity in human cortical organoids ,and hippocampal long-term potentiation in vitrified and rewarmed adult mouse hippocampal slices . These protocols are just beginning development, and not many labs are yet working in this field - we don't yet know to what extent protocols could be optimized to protect neural tissue.
To give another objection - one might assume that the brain would simply be unable to re-start after such an intervention. And that is certainly a possiblity - we won't have more information until we try these experiments in preclinical models. But, surprisingly, it's actually common practice to flatline patients on EEG with induced hypothermia in order to extend time for surgery. People are regularly cooled to 18°C during aortic arch surgery, slowing biological time so dramatically that they flatline on an EEG, yet are rewarmed and regain normal function . While this isn't vitrification (it's specifically an example of hypothermia-induced flatlining, not functional recovery of neural tissue from vitrification protocols), it—along with anesthesia—is a surprising example of the brain recovering from non-physiological changes of state which disrupt neural activity. The true test, however, will be first taken in preclinical studies, such as seeing if we can wake up rats from cryopreservation, and preserve memories and social behavior encoded prior to cryopreservation.
I believe that we should run these questions down with the full force of our current scientific armory. I’ve bet my career on it, because it feels viscerally important - the clearest problem to work on I’ve met in two decades of screening technical prompts in biotechnology. We don’t know the answers to some of the questions involved, but this question is crying out for focused, large-scale engineering efforts. After all, lives are at stake.
This table gives concrete examples of a general pattern: in many years, large numbers of people die from diseases that, within the next 1–2 years, become much more treatable or even curable. Selected conditions with clearly identifiable “step-change” therapies (e.g. HAART, DAAs for hepatitis C, PD-1 inhibitors, CFTR modulators), substantial global burden, and reasonably accessible data on annual deaths and treatment impact. For each disease we chose a “just-too-early” year, typically just before regulatory approval or publication of pivotal trial results for a transformative therapy, and treated that moment as a simplified step change, even though in reality uptake is gradual.
Estimated “potentially benefiting deaths” start from global deaths from that condition in the chosen year and, where appropriate, are adjusted by subtype, stage, biomarker status, or genotype to approximate the population that would later meet something like trial-eligibility criteria. These are deliberately rough, back-of-the-envelope figures, intended to illustrate scale rather than provide precise counts of lives saved. They do not assume everyone counted would have received the therapy: access barriers, comorbidities, performance status, competing risks and trial-like eligibility constraints all reduce real-world impact. The numbers should therefore be read as an approximate upper bound on people who plausibly could have benefited. We use a 12–24-month window for simplicity and conservatism; for many chronic diseases, a longer (5–10 year) horizon would substantially increase the number of people who might ultimately have been helped.
Even with conservative assumptions and real-world constraints, we can see that in some years on the order of hundreds of thousands of people died from diseases that soon became much more treatable.
Why might reversible whole-body cryopreservation not be the best option for a patient? We can imagine patients who, in advanced stage of sickness, would not be the best fit for a major surgery. Patients who highly value time with their family and loved ones, and wouldn’t want to tradeoff a chance of longer life with them later, for more time with them now. Patients with diseases which, for some reason, we have no hope of curing. Patients who die outside of a hospital setting, and for which a substantial risk of death could not have been predicted. Patients with advanced dementia might not find their identity recoverable, and therefore quality of life sufficient, even if lifespan was increased.
Jump to figure ↑Dying on the brink of a breakthrough
In the past decades, there have been years in which anywhere from 150,000 - 835,480 or more people (0.27 - 1.4% of all deaths) have died from diseases that received a life-prolonging therapy within the following one-to-two years.
To explain why my co-founder Hunter and I care about reversible whole-body cryopreservation, I’ll share a story in his words. “In 2016, my father-in-law, Mark, was diagnosed with a terminal case of mesothelioma and died months later after frontline therapies failed, as we knew they would. Near the end of his life, a Keytruda trial opened for his disease, but he was too sick to qualify. Keytruda is now recognized as an effective therapy for some mesothelioma. When we heard from his doctor that there was nothing left to do, I felt a profound powerlessness. For all my understanding of medical physics and biology, I couldn’t help a man I loved.”
| Disease / condition | Break-through that arrived ≤ 12–24 mo later* | Typical survival gain vs. earlier standard | Deaths in the “just-too-early” year | Share of all global deaths¹ |
|---|---|---|---|---|
| Metastatic non-small-cell lung cancer (NSCLC) | 1st-line PD-1 inhibitor pembrolizumab (US approval late 2015) | Median OS roughly doubled in PD-L1-high tumours (∼13 → 26 mo); 5-yr OS estimated 32 % vs 16 % | ≈ 1.69 million lung-cancer deaths in 2015 , ~87% of cases are NSCLC (an imperfect approx for % of deaths) , ~44% of relevant diagnoses were Stage 4 in 2015 , 23.2% have PD-L1 in at least 50% of tumor cells -> approximately 150,000 deaths |
≈ 0.27% (150,000 / 56.3M deaths in 2015 ) |
| Advanced melanoma | CTLA-4 antibody ipilimumab (2011) followed by PD-1 combos | Ipilimumab + gp100 in patients with unresectable stage 3/4 melanoma increased survival to 10 months, from 6.4 months with gp100 . Ipilimumab combined with nivolumab later led to median survival of 71.9 months, with a melanoma-specific survival rate of ~52% at 10 years. | ~49,100 deaths from malignant skin melanoma in 2010 | 0.09% (49,100/54.3M deaths in 2010) |
| Cystic fibrosis (~90% of patients) | Triple CFTR modulator elexacaftor/tezacaftor/ivacaftor (“Trikafta”, 2019) | Significant FEV₁ improvements; projected extra life decades ; registry already shows falling mortality rate . Among 65 lung transplant candidates, after therapy, 61 no longer met transplantation criteria . | Mortality rate ~1.3 for per 100 people with CF in 2018 in the Cystic Fibrosis Foundation registry , with a global estimate of 162,428 CF patients around 2020 , so possibly ~2,112 deaths, of which ~90% might be eligible for CFTR modulators. So, up to ~1,900 deaths. | 0.003% (1,900 / 57.8M deaths in 2018) |
| HIV infection (historical example) | 1996 - First triple-combo HAART efficacy readouts | AIDS deaths fell 60–80 % within 2 y in treated regions | ≈ 1,090,000 AIDS deaths in 1995 | 2.1% (1.09M / 51.4M deaths in 1995) |
| Hepatitis C | 2013 - FDA approves sofosbuvir, an oral NS5B inhibitor that enabled widely used interferon-free HCV regimens with ≥90–95% cure (SVR) rates ) | Functional cure = near-normal life expectancy | ≈ 479,381 k HCV-related deaths in 2013 | 0.87% (479,381 / 55.1M deaths in 2013) |
Interestingly, at the temperatures achieved in vitrification, patients can be stored for (theoretically) for up to millions of years - whole human embryos have already been stored for decades, thawed, and used to create healthy babies. If we widen the window from 1-3 years, to 5-10 (still, possibly, within the window of acceptable social change for most patients with terminal illness), the numbers grow even further - for NSCLC lung cancer alone, 750,000-1 million people could have been helped.
In medical technology, anti-memes cost lives
Reversible cryopreservation was, for me, an easy idea to dismiss, despite a two-decade hunt for high-impact ideas in medicine. In all of that time, I never seriously considered working on it. It didn't seem plausible - I had some hand-wavy ideas about ice formation being an issue, but mostly, I just kept running into easy reasons to dismiss it and didn't put in the time to research them a few more steps. I’m worried that this blindspot isn’t unique to me, and that the field of cryobiology is systematically missed by operators, scientists and funders who should be working on it.
It's an excellent anti-meme - an idea good at hiding from those looking for it. You can't grasp it from one angle - it requires juggling evidence from biochemistry, thermal engineering, materials science, clinical practice, and many other domains all at once. In some contexts, specifically whole body work is outside the Overton window of current scientfic convention - I know anecdotally of grants for whole-body work in insects being dismissed as 'too risky' because they included a reference to 'whole body' work, or even the word 'hibernation' being too risky to include in a grant proposal. So social heuristics might indicate one should not work on it.
I can see this bias in my own thinking - on some level, I was busy analyzing a lot of ideas, and it was easy to find quick objections that meant I didn’t have to think about it critically. I batted it away with: ‘it would scramble molecular information!’ or ‘tissue is way too complicated!’ or ‘you can’t expose cells to such high chemical concentrations in an actual clinical context!’. For years, I failed to update on the easily available evidence that research labs regularly cryopreserve a diverse array of cell types for long-term storage, small mammalian organs had been reversibly cryopreserved and clinicians regularly opt to expose human embryos to high concentrations of cryoprotectants for short time periods during IVF. But also, none of my friends or colleagues were working on or talking about this field - it just didn't enter into set of ideas I took seriously because I looked around me and saw people I knew and respected working on them.
I certainly don't think we can declare the problem provably solvable now. But I believe it's a problem worth working on. My co-founder's father-in-law died because he was a few months late for a critical trial. In some years, over a hundred thousand people miss a life-saving therapy. By a matter of months. Transplant patients waiting in the hospital miss organs that expire in transit. We need to urgently investigate whether fundamental barriers to scaling cryobiology protocoles exist, and if they do, hammer away at them to see if any cracks can be found. Given the lives at stake, I believe we must dedicate an order of magnitude more funding and intellectual talent - if not two - to developing and translating cryopreservation protocols.
But that isn’t the only way in which working on this problem changed me.
Philosophically, the best arguments for continuity I have right now are that the authenticity of an object might count for something (if you know it had a history, you can infer things about it and the world which you couldn’t without that knowledge) - something like the authenticity of a soul. In addition, there is some idea of a garden aesthetic - growing an emergent, complex system over time - that feels slightly at odds with a rapidly duplicating, replicator-driven future, and might lead one to have a preference for some kind of continuous and less well-defined process.
Jump to figure ↑Part 3: Identity, and how spending time with this idea changed me
I wanted a technology which, if built, would give someone I love a fighting chance if they were on their deathbed.
This was, initially, my pull to the problem. I was afraid of dying, and thought that working on such a technology might save me and my loved ones from death.
But, in working on it, I changed, in ways I am still coming to terms with.
Cryopreservation forces deep questions about identity. When someone goes to sleep, are they the same person when they wake up? What about someone who undergoes anesthesia for surgery, or temporarily flatlines on EEG during deep hypothermia-induced cardiac arrest? What about someone who is cryopreserved and then rewarmed - are they the same person when they wake up, if they act the same?
To make these questions concrete, let’s walk through the teleporter thought experiment, posed by philosopher Derek Parfit. Imagine a teleporter offers to send you to Mars. At first, you're excited. Then, you learn its method: it creates an atomically perfect copy of you on Mars while instantly and painlessly destroying the version of you on Earth.
This poses the essential question of continuity: Is that person on Mars really you? Or are they actually a clone — an utterly different person, simply because they are not continuously connected to your consciousness?
Before working on this problem, I would have found stepping in the teleporter incomprehensible. I couldn’t even describe why, it just felt absurd. But in working on the problem of cryopreservation, I had to compulsively examine my beliefs about identity.
Protocols which preserve the option for continuity are more difficult to develop than those which don’t, and I cared about working on this technology to preserve life for those I loved so it was important to define what it was that I cared about. I entered a state in which every day my brain would loop on the question ‘how am I compared to the person I just was a second ago’. It was extremely disorienting - some days, even getting lunch with a friend felt like an intense, disorienting experience, when my brain couldn’t put the question down in the background.
It turns out (though I didn’t know it when I started) this practice (constantly asking yourself “what is the self? How am I the same person that I just was?”) looks a bit like what you do when pursuing a Buddhist meditation on the concept of no-self. I spiralled, and for several months, there were days when every second I felt as though a version of me had ‘just arrived’, out of a void, into that moment. It was a strange time. And at some point, the bottom fell out of a core belief I’d had, and I lost track of something I had previously been quite confident about. I felt that the person I was in a given second, and the person I was in the next second, were similar, but the fact that they were continuously connected wasn’t meaningful to me in the deeply visceral way it had been. And the fear of death that, on some level, I’d defined myself by publicly and internally for two decades wasn’t there anymore.
It was, again, a very strange experience. On the one hand, I now understood, at least to some degree, why someone wouldn’t get in the teleporter. I noticed that I felt a lot more likely to want to help or be compassionate toward people who weren’t me - I felt logically as though I should care about other people, literally, as much as I cared about myself, or care about supporting their work to the degree that we shared values or goals. On the other hand, the core logical motivator for my work to date disappeared (if continuity of consciousness wasn’t important - was anything in medicine? Presumably some continuity was important - otherwise we’d all just be babies unable to do anything, and people doing things seemed good? But was it better to just focus on more people existing overall, doing ‘good’ things, than on specific people not surviving terminal illness? How much continuity was important in the abstract, if it wasn’t in of itself a goal to be pursued?).
It took a few years of what felt like constant questioning, but the state eventually settled back to a philosophy I can live with - and which likely will continue to evolve over time. To be honest, it happened when I had a dream about someone I love a lot facing death, and how they didn’t want to and it wasn’t the right time for them. I woke up crying, and realized that I’m honestly just still deeply compelled by what feels ‘right’ for people I care about, and for a lot of them, the current state of medicine is horrifying. It might be possible for us to, personally, overcome a fear of death - and perhaps it’s a philosophically preferable position for some - but many people, obviously, die sooner than they would like. Grandparents who wanted to see their grandchildren grow up, or people working on projects they cared passionately about and didn’t get to see all the way through.
One of my favorite directors, Satoshi Kon, died of pancreatic cancer at age 46, midway through completing his final film, ‘Dreaming Machine’. The storyboards were complete, but the full film will never be finished. I watch his films every year, and wonder what new worlds he would have created if he’d had the time he needed.
More poetically, I want to say something like - what kind of art could we make if our minds had much longer to evolve? Take the Great Wave off of Kanagawa, one of the most reproduced and compelling images on earth, created by Hokusai at age 71 - a consciousness that had a long time to evolve.
“From the age of six, I had a passion for copying the form of things and since the age of fifty I have published many drawings, yet of all I drew by my seventieth year there is nothing worth taking into account. At seventy-three years I partly understood the structure of animals, birds, insects and fishes, and the life of grasses and plants. And so, at eighty-six I shall progress further; at ninety I shall even further penetrate their secret meaning, and by one hundred I shall perhaps truly have reached the level of the marvellous and divine. When I am one hundred and ten, each dot, each line will possess a life of its own” - Hokusai, Calza 2003
I would have loved for Hokusai to have as much time as possible to create art which had a life of its own.
Imagine a future in which spirit floats in a way that is less constrained and more empowered. While acknowledging the natural flow and ebb of life - we can make a new, evolving, and more beautiful tapestry, and should be open to helping those we love evolve into the most full and empowered versions of themselves
Cryo is, strangely, the most conservative of all transhumanism technologies (in a sense) - it is deeply hopeful about humans continuing to exist in their current form. It wants them to travel through space and time, as they are, going to the frontiers of the universe. Cryo deeply wants human bodies to be part of the far future - we are humans, in bodies, and perhaps should want this also.
I wanted to write this piece to explain something which I think is important. But also, I wanted to externalize these aspects of an idea which I find deeply beautiful, which fascinates me, and which continues to change me and influence me the more time that I spend with it.
Ideas, to me, feel like objects with their own soul and structure. I can feel them nearby - glistening, beating, moving with intention through multiple minds.
I’m not the same person I was when I first met this problem - the future feels fuzzy, and hazy, but the problem still feels deeply beautiful and beckoning. It calls to my mind, to show me new delights every time light glints off another of its many facets. And perhaps now you will say hello to it as well.
Appendix
Some molecular anecdotes that didn’t quite fit in the main text, but I’d be loath to part from
Cells can survive randomization of molecular motion - why?
Imagine standing inside a cell - you'd be bumped a trillion times by molecules dancing their thermal ballet. Now imagine freezing this dance mid-performance, blindfolding every dancer, spinning them around, and yet somehow when they begin moving again - choosing random directions - they still dance in perfect synchrony. This is essentially what happens in reversible cryopreservation.
And it doesn’t break the system! Why on earth might this be the case?
The short answer is (at least in part) that we evolved to tolerate thermal noise.
Being in the cell is like being in the middle of a crowd sprinting into a packed stadium. You’d quickly get bumped around, even trying to get to your seat - it’s not realistic to run through a packed, jostling crowd in a straight line. Similarly, cells don't try to fight the chaos - they embrace it. Without this property, reversible cryo would be impossible.
The cell treadmill experiment
One thing I enjoyed thinking about is, to the above thought experiment - how does the cell not only tolerate, but take advantage of thermal motion? This got me thinking about a cell treadmill - what if the cell didn't have thermal motion, and had to ferry molecules around in some mechanical way to get to the right place, when they otherwise didn't move?
Let’s do a (very hand wavy) thought experiment, where we try to imagine the kind of crazy dynamics a cell would have to put together if it wanted to run itself more like a human factory, and less like a passive diffusion computer - if a cell gets stuff to happen by bumping molecules into each other, how else might they do that? Imagine the counterfactual - what if instead of them just constantly bumping into each other, cells put molecules or proteins on little treadmills and moved them between other waiting molecules or proteins. If there are, let’s say, 10^9-10^12 reactions / second, and let’s say a treadmill is ~100 atoms by ~100 atoms, and the length of the cell (accounting for some space) - you could pack 100 high and 100 wide in an E. Coli. So, 10,000 treadmills, ~10,000 atoms long. Let’s say a substrate is ~100 atoms across - so you can fit ~100 substrates across one treadmill, and you have 10,000 treadmills, so 10^6 substrates. The treadmills would still have to run at 10^3-10^6 * 100 atoms / second, so 10 microns - 1 cm / second. The fastest that a motor protein scoots things around in a cell is ~1 micron a second, so this might be a bit out of reach. And that’s assuming their ordered such that the substrates always meet the right protein next, which they wouldn’t in this context! Much the less the energy that would be needed to fund them if you didn’t have temperature. Randomly bumping into everything saves you a lot of organizing, in that context.
Also, it’s super weird that the energy of a random bump is similar to the energy needed to do stuff (despite the two seemingly being determined by different physical constants). This paper from Ron Milo and Steve Quake is super interesting in this respect!
Along these lines, it's also interesting to read, in Schrodinger, a description of the beauty of biology - linking 'interesting' phemonena at atomic scales with macro-scale statistical properties
Notes
Notes on Figure 1
This is a graph I just drew to illustrate the point, but it’s not totally arbitrary - it’s inspired by the graph below (adapted from the Cystic Fibrosis Foundation), which shows a surprisingly monotonic upward trend in life expectancy - and that’s not counting the (estimated) ~2 decades of additional lifespan from the most recent 2019 approvals .
It’s also interesting to me that life expectancy looks (for the most part) upward monotonic, even when counting for the effect of decreased child mortality with time , and across countries .
Notes on Figure 2
Cell counts were inferred as follows:Sperm cells are representative of single cell cryopreservation.
For blastocysts, the Cohen et al citation did not report the number of cells in the implanted blastocyst, but blastocysts were reported to reach on average ~58, 84 and 126 cells on days 5, 6, and 7 so I used oom 10^2 cells.
For C. elegans, the number of somatic is exactly known - 1031 (although possibly this number should be updated to 1033) for mature males, 959 for mature hermaphrodites . So I used oom 10^3 cells.
For rat kidney, an exact cell count would not be currently possible, so we can check two different methods to see if they align.
First, find a lower bound by # of glomeruli, and # of cells per glomerulus. From , we see that an adult rat Sprague-Dawley rat kidney has 31,764 +/- 3667 glomeruli, with each containing 674 +/- 129 cells. So, we can assume a lower bound of 674 * 31764 = 21,408,936 (oom 2x10^7) cells. But glomeruli are just a subset of kidney tissue, so the actual number might be much higher.
Now, let’s check a naive other estimate - if we divide the weight of an average kidney by an example mass of a mammalian cell. The mass of a single HeLa cell was estimated ~1.01-3.57ng . Mean wet kidney weight at 13 weeks of age was estimated at ~1,300mg . Taking the (rough) ratio of the two, we might estimate 1.3 g / 10^-9 g ≈ 10^9 cells, or 1 g / 4 x 10^-9 g ≈ 2.5 x 10^8 cells. But these are wet mass ratios, and could be confounded by non-intracellular water in the kidney - re-doing the calculation with dry mass, we see the mean dry mass kidney weight at age 13 weeks was estimated at ~318 mg , and reasonable heuristic for cells is ~70% water by mass , so we get dry mass of a HeLa cell at (roughly) .3 * 1ng to .3 * 4 ng = 3x10^-10 g to 1.2x10^-9g . Dividing kidney dry weight (rounded form 318 mg to ~3x10^-1 g) by these numbers, we get 3x10^-1 g / 3x10^-10 g to 3x10^-1 g / 1.2x10^-9g ≈10^9 - 2.5x10^8 cells.
So, we might guess oom 10^9, with the possibility for error if the acutal number is closer to 10^8.
For human kidney, we can do a simple mass ratio check - for a 130-170 pound man, median right kidney weight would be 140-150 grams . This is ~2 orders of magnitude higher than rat, so we might naively guess a correspondingly higher cell # at 10^11. To quickly gut check this, if corresponding total human cell number is 3x10^13 , kidney cell # would represent ~0.3% of the cells in a human. The ratio of kidney weight to total body weight from the above numbers could be estimated at ~150 g / ~70 kg -> ~.2%, so 10^11 does not fail a basic gut check. That said, I think RBCs dominate the 3x10^13 cell count, and are quite small, so there’s a chance I’m still overcounting.
Lengths for Figure 2 were inferred as follows:Sperm cells - I used human as they were tested (alongside other species) in the cited study. They are cited as 5-6 micrometers long, and 2.5-3.5 micrometers wide - taking the largest dimension cited (length), I rounded up to 10 microns (which is also pretty representative oom for mammalian cell size).
Blastocysts - From , day 5 blastocysts were measured at ~160 microns (summarizing data from Table 1 of the paper), so I used oom 10^2 micron.
C. elegans - Adult hermaphrodites are ~1 mm long (see figure 6).
Rat kidney - In Sprague-Dawley adult male rats, inner medulla was measured to be 4.5-5.5 mm, and outer medullary thickness was 2-2.2 mm, so I used 1cm as an oom for kidney width
Human kidney - From , the median renal right kidney width measured in male volunteers was 59 mm, and median length was 112 mm, so I rounded up to use 10 cm as an estimate.
Note on calculating the # of atoms in an E. Coli cell
Reference estimates 10^10 carbon atoms in the dry weight of a medium sized E. Coli cell. According to the ratio given (C4:H7:O2:N1 or normalized to carbon C:H1.77:O0.49:N0.24), for every carbon atom there are ~2.5 other atoms in the dry weight portion of a cell, so we might estimate ~3.5 x 10^10 atoms from dry weight. To add the wet weight component, we can first multiply each atomic component of the dry weight be the atomic mass (so, 4x12 + 7*1 + 16*2 + 14*1 = 101 Da) - for 10^10 carbon atoms, that would be (correcting for carbon 4x in the ratio) ¼ x 10^10 x 101 Daltons in dry mass ≈ 2.5 x 10^11 Daltons. Given water is 70% of the cell by mass we multiply by 0.7 / 0.3 to get ≈ 6 x 10^11 Daltons, and divide by the atomic mass of water (≈18) to get ≈ 3 x 10^10 water molecules, or 9 x 10^10 additional atoms. So overall, adding dry and wet weight atom numbers, ≈ 1.25 x 10^11 atoms.
Notes on the 'danger zone' for ice formation
The ‘danger zone’for ice formation is characterized as starting below 0C , through to the glass transition temperature (typically around -130C) at which point viscosity increases to the point where there is essentially no molecular motion to cause ice formation.