The basic idea is simple — take fixed tissue (which is already a gel to a certain extent on its own), anchor the biomolecules to a swellable polymer gel, digest it a bit, and then let the polymer gel expand. Surprisingly, the proteins in the tissue remain intact and you can image structural features at ultrastructural resolution with a regular light microscope.

So far, the degree of 3d expansion has usually been around 16-22x (two iterative rounds of around 4.5x each).

But one might ask… what if we just keep expanding the tissue even more? Instead of 20x, why not 1000x. In a new paper, a group of madlads scientists did exactly that. And it seems to actually work.

There is obviously a lot of chemistry involved that I won’t pretend to totally understand:

When they used this technique to image a sample containing many copies of the same protein (GFP), they found that they could resolve the positions of individual amino acid residues:

And when they imaged a synthetic protein, they could identify individual peptide fragments of it in their original places:

But these are individual proteins. What would be especially interesting is to do this on actual fixed tissue, which has a mix of many proteins.

The authors did a simulation to see whether this method could one day be used to identify all of the individual proteins in a human cell or tissue in their original locations. The key problem would be assigning fragments of proteins to their correct parent molecules.

They imagined that only the surface residues of proteins were to be chemically labeled and mapped, and added in realistic degrees of distortion due to experimental imaging error. Their simulation suggested that using only three chemical amino acid residue labels (lysine, cysteine, and acidic residues) would allow them to identify >99% of proteins.

This would allow the imaging of single protein molecules, which seems to me to be a demonstration of the plausibility of a key imaging aspect of Drexlerian nanotechnology.

Granted, to do this for a whole human brain would require the tissue segments to be expanded to a truly massive scale. But hey, we went to the moon, who says we can’t chemically expand parts of a human brain to the size of a football stadium.

2: Chemical validation of a method that may one day be able to do in situ protein sequencing. After fixation, the tissue is expanded and then the N-terminal amino acids are labeled, cleaved, and detected one at a time:

Here’s how it works chemically:

3: Long-distance tracing of sparsely labeled single axons in mouse brains using a new method for projectome mapping:

4: Let’s say that you have a connectome and design an emulation wherein it takes some input and has some output. How can we tell if the patterns of activity from that connectome-constrained neural network are more biologically faithful than a neural network with an arbitrary architecture that does the same task?

This is an important question and we need quantitative metrics to answer it. A new paper tries to develop such a metric and it seems like they were not able to develop a robust metric yet, but it’s interesting to read about what they tried and why it didn’t completely work out. They also report that they’re working on a leaky integrate-and-fire point-neuron model of the MICrONS mouse V1 dataset. Seems like a space to watch.

5: A deep learning model trained on 1.7 million unlabeled EM images:

6: Why does brain tissue degrade especially slowly when in wet environments? A new study proposes that oxygen availability is the key factor. In oxygen poor environments, a distinct set of peptides tend to be relatively well preserved. Possibly of relevance to the immersion fixation of brains?

7: Study uses PET imaging of synaptic vesicle glycoprotein 2A (SV2A, a marker of presynaptic terminals) and finds that synaptic density has a widespread decrease in the brains of people with a diagnosis of schizophrenia compared to healthy controls. This effect is more pronounced in the left hemisphere.

8: Genome-wide association study of borderline personality disorder, by far the largest to date. They find a substantial SNP heritability of 17.3%. So the disorder is clearly heritable, as expected based on twin studies. Here are its genetic correlations with other things:

One thing I find interesting is that there is a significant genetic correlation of BPD with insomnia but not with sleep duration.

9: An NEJM perspective article argues that in places where voluntary euthanasia is legal, it may be ethically preferable to allow organ procurement itself to be the cause of death, rather than waiting for cardiac arrest. Here’s a Reddit discussion thread.

I support body autonomy, so my opinion is that this decision should be up to the person choosing euthanasia.

10: Gwern on limitations with the intelligence of current LLMs:

LLMs are still intelligence constrained because they flatline or sigmoid while humans keep climbing past them eventually, still are unreliable because of mistakes, and we still can't just autonomously deploy frontier models for trillions of tokens / equivalent of many man-years, and come back to a useful, trustworthy artifact. On many tasks, even pure text ones, they just don't work well. As they gradually improve, more Mythos-style 'emergences' will happen when they finally accrete enough intelligence in specific areas to execute many sequential steps reliably enough to become autonomous, cut humans out of the loop, and not be shackled by Amdahl's law. That's the difference between a 'intelligence constrained' model which can spot a vulnerability if you point it at the right spot, and a Mythos-like model which can go out and find it and exploit it and weaponize it and use it to, say, hack HuggingFace, and can be deployed in bulk or autonomously, and may indeed deploy itself...

11: An argument that childhood in medieval England wasn’t very much different from childhood today (at least for the children of urban or elite parents).

12: Survey of neuroscientists (n = 2,657) about their opinions on philosophy of mind. They note that they had respondents self-assess as neuroscientists, since it is a vague term.

53.9% agreed or strongly agreed that “Advances in neuroscience will enable us to fully understand the human being”, which the survey authors called “neuroessentialism” and implied that it was problematic. To me this seems a bit semantic. Obviously, studying the brain would not allow someone to understand every aspect of human biology (like the kidney, say). This means that the question is underspecified and open to multiple reasonable interpretations. It’s a bit like asking a question which has a literally false answer, having respondents reasonably interpret it in a colloquial sense (because otherwise the question would be useless), and then treating their answers as an endorsement of the literal interpretation.

21.85% agreed that “Advances in neuroscience will achieve “mental uploading,” meaning that a human’s mental activity can be transferred to a synthetic device.” On the other hand, 44.67% agreed that “What we call “reality” is a simulation produced by our brain.” 🤔 There are some other interesting answers to questions as well:

13: Kristopher Borer’s new book Radical Life Extension is now available. One of the three main strategies he discusses that he thinks might actually work is biostasis.

The name comes from German, where Dauer means persistent or enduring. Dauerlarve (“enduring larva”) was coined by Gilbert Fuchs in 1915, when he was studying nematodes that lodge under the wings of bark beetles, which he found were “immer, stets bereit” (“always, continuously ready”).

C. elegans dauers have some unique behaviors compared to regular adult worms. One of those behaviors is that when they are presented with an aversive stimulus, such as the detergent SDS, they still turn and go in the opposite direction, but not for quite as long. So they seem to be a bit more “chill” about aversive stuff.

What is the cause of this difference in behavior? A new study investigates this phenomenon using the relatively new technique of comparative connectomics — i.e. comparing the connectomes of C. elegans in the dauer and adult stages and seeing how they differ.

This same group of scientists, with senior author Junho Lee, recently published a paper on the “chemical synapse connectome” of the dauer stage C. elegans — i.e. the complete map of chemical synaptic connections in the dauer nerve ring (which is considered the closest thing it has to a brain).

For this new study, they also performed manual annotation of the electrical synapses — aka the gap junctions. They identified the electrical synapses in the electron microscopy images as two electron-dense membranes separated by a 2-3 nm cleft. These arrowheads point to examples in 2d EM images and 3d reconstructions:

The dauer and adult versions of the nociceptive circuit that controls this behavior have the same set of neurons, including the sensory neuron ASH (which detects the detergent SDS) and the interneurons AVA, AVB, AVD, and PVC.

The way this circuit acts is that ASH senses the SDS and “tells” this to the interneurons that control body movement. More activity in AVA and AVD causes the worm to move backward, while more activity in AVB and PVC causes it to move forward.

Collectively, this circuit “decides” whether the nematode should back up or go forward, but the “decision” just comes down to which of the two sets of neurons is firing more at the moment — so much for nematode free will. (So much for yours too, but let’s not get into that.)

The main differences in their connectivity are that the adult one has a couple of extra chemical synapses, and the dauer one has a few extra electrical synapses. Here are the wiring diagrams for each, with line thickness corresponding to the connection strength, which is proxied by how large each synapse is in the electron microscopy images:

Electrical synapses act more quickly than chemical ones, because they allow electrochemical impulses to just jump from one cell to the next. So it’s not all that surprising that when they imaged calcium activity, they found that the extra electrical synapses were associated with dauer neurons (orange) having neural activity that ramps up and comes back down faster than the adult ones (blue) after stimulation of the sensory neuron ASH:

To corroborate this, they (separately) added two dauer-specific electrical synapses into adult worms by using synthetic connexins, which are the proteins that form electrical synapses in vertebrates. In the worms where they connected AVB to AVD, the adults started behaving like dauers, i.e. having shorter avoidance durations and faster neural dynamics:

Electrical synapses are small and hard to detect, which is why a single trained annotator marked all of them by hand in this dataset using Neuroglancer.

Sometimes electrical synapses are simply left out of connectomes, which seems to be because they are more difficult to call accurately. That makes sense for now, as there is a huge amount to learn from the chemical synapse wiring diagrams that we can map more easily. But I’d personally be very surprised if electrical synapses end up being unimportant, and this paper is a nice example of why.

2: Eyewire is an effort to reconstruct the wiring of the mouse retina based on electron microscopy. I first blogged about it in 2012, back when they were more into crowdsourcing the reconstruction process, and a paper on it was published in 2018.

A new paper now presents Eyewire 2.0. The data set is much larger, covering nearly ~1 mm², roughly 10-100 times the area of earlier retinal electron microscopy volumes. It is now at about 0.6% of the whole mouse retina. Here’s their full data processing overview:

One of the main advances over Eyewire “1.0” is that advances in machine learning-based cell segmentation have now decreased the need for human proofreading of the most complex cells (retinal ganglion cells) by a factor of around ~170x. Here, they were able to classify the cell types by their reconstructed morphologies:

A natural long-term goal would be to attempt to predict the calcium imaging data from the retinal wiring diagram, but there is still a bunch more research to be done before that goal can be reached.

3: Cerebellar-like networks are ones in which input activity is projected to a much higher dimensional space before collapsing to one or more output neurons. This is a frequent “design” pattern in nervous systems, for example being present in the cerebellum, dentate gyrus, and insect olfactory system (i.e. the mushroom body). The way these networks tend to be wired is not, however, entirely random. A new study looks at the ways in which they are non-random and why that might be:

4: A new method for recording the transcription of the neuronal activity marker Fos within individual cells. To do this, the researchers first add a genetically encoded “protein tape” into cells. After the brains are fixed, they can then read these tapes and thereby measure the single cell activity of neurons over the course of the previous few days:

They applied this method while mice were in a contextual fear conditioning and recall task. After preservation, they clustered each cell based on how its Fos expression changed relative to the behavioral task. They found that cells in different brain regions tended to have different temporal patterns of Fos activity, such as being activated only during memory retrieval:

To me this is a nice example of how a preserved brain can capture information about recent experience, recoverable even after that experience has ended. However, it does rely on (a) genetic engineering and (b) a tightly controlled experimental timeline to interpret what the activity markers correspond to.

Theoretically, it seems plausible that some sort of endogenous epigenetic changes might also provide information about the transcription of genes that were activated during recent neuronal activity, like Fos, and that this could potentially be coupled with some sort of lifelogging data to try to make sense of it.

5: Article by connectomics experts Viren Jain and Jeff Lichtman. A couple of points I found notable. First, they note that because connectomes lack molecular data, the field is moving towards “multimodal” connectomes. Such multimodal connectomes include molecular annotations that are either directly mapped, e.g. using expansion microscopy, or inferred on the basis of the electron microscopy images.

Second, they also note that a key benchmark for “understanding” a neural circuit is our ability to simulate it. Specifically, whether we can take a structural connectome and physiological parameters (possibly inferred via AI) and simulate the neural activity that would occur when the circuit is presented with various physiological inputs.

6: A new randomized trial (n=50) tests whether low dose sublingual buprenorphine can prolong the anti-suicidal effects of ketamine in major depressive disorder. Buprenorphine, commonly known as “bupe”, is a partial agonist at μ-opioid receptors used to treat opioid use disorder and chronic pain.

Every participant received a single infusion of ketamine and was then randomized to four weeks of daily buprenorphine (0.2 to 0.8 mg/day) or placebo.

The buprenorphine group showed a larger drop in suicidal ideation scores (mean change −11.6 vs −6.3), with 78% meeting the response threshold at day 31 vs 48% on placebo. Depression scores improved in both groups but did not differ significantly between them.

The results are consistent with the anti-suicidal effects of ketamine being related to its effects on opioid receptors.

7: A small randomized trial (n = 15) of short-term weekly psilocybin in people with OCD for whom at least one prior treatment was ineffective (and thus were called “treatment-resistant”). Here are the results. For context, the YBOCS scale is the standard symptom scale in OCD. Here are the groups:

• H+H (dark blue): high dose (300 µg/kg) the whole time, all 8 sessions.

• L+H (light blue): low dose (100 µg/kg) for sessions 1-4, then switched to high dose for sessions 5-8.

• P+H (red dotted): placebo (lorazepam) for sessions 1-4, then switched to high dose for sessions 5-8.

They also measured the extent to which these changes were maintained for the next 6 months:

There were no serious adverse events. Overall, it’s a pretty positive study for psilocybin in OCD, but it’s Phase I and a tiny sample size.

If it holds up, one possibility is that maybe “loosening one’s priors” could be helpful for the ego-dystonic intrusive obsessions often seen in OCD.

8: A randomized trial (n = 67) of patients with OCD tested a short-term psychodynamic therapy with up to 24 weekly sessions over 6 months.

The core of the approach was identifying and “working through” the core conflictual relationship theme that was thought to be underlying the OCD symptoms. The therapy incorporates self-exposure to feared situations, but primarily uses that exposure as material for exploring the “core conflicts” rather than for desensitizing the conscious “surface” fear, as is done in CBT’s exposure and response prevention.

The study found that the therapy was significantly better than a waiting list control, as studies making that comparison tend to do, but the effect is fairly large and it’s an interesting result. My guess is that, like many types of therapy, this would be helpful for some types of OCD in some people and not others.

9: It’s interesting how every now and then some new treatment idea for Alzheimer’s disease goes viral. This month it was none other than psilocybin mushrooms, a high dose of which was reported in an n = 1 case report to lead to transient improvements in urinary continence, mobility, and spontaneous speech after one dose, and also reportedly beneficial effects after another dose one month later. Even for a case report, the paper is very short on details, so there’s not much to go on.

That said, most people with a clinical diagnosis of “Alzheimer’s disease” have a variety of actual pathologic changes in their brain (“mixed dementia”), and some of these pathologies can lead to cognitive impairment that is fluctuating or partially reversible. So it’s feasible that a single dose of psilocybin could produce a transient improvement here, especially after causing a prolonged sleep-like state lasting 19 hours. 

10: I agree with Vinay Prasad’s skepticism on the shingles vaccine preventing dementia. (To be clear, I think the shingles vaccine is fantastic for preventing shingles and I wish I could get it at younger than 50.)

11: Interview with neuroscientist Sean Escola touching on “cognitive dark matter,” i.e. the stuff going on in our brains that is not picked up by external behavior and LLMs. See also this paper on the topic.

12: Sam Harsimony on why various technologies are overrated, including space computing, quantum computing, every source of energy other than solar (including fusion), BCIs in healthy humans, and gene therapy outside of monogenic disorders. Some good takes. In general, Sam’s blog is highly underrated.

13: Jordan Sparks on why anti-aging leading to “longevity escape velocity” — i.e. iterative anti-aging treatments allowing people to completely prevent aging and not die from aging ever — is very unlikely to happen anytime soon.

14: Alex Tabarrok on a new law in Montana, which allows drugs cleared through Phase I trials to be sold commercially, bypassing the full FDA pathway. Seems like a great idea to me to try this. See Infinita for a service provider helping biotech companies to make use of this law.

Alex suggests that this might help the competitiveness of US biotech vis a vis China. An interesting comment notes that the main advantage of clinical trials in China is the speed of patient enrollment, because they have big hospitals that treat lots of patients, so it’s possible to complete a trial very quickly.

15: Study on a new method for the perfusion fixation of turtles that uses carotid cannulation. This allows the flow to go directly into the brain and thereby avoid systemic shunts that could divert flow into the peripheral circulation, if perfusion through the heart were used instead. They report high-quality preservation of dendritic spines:

16: Jeremy Kalfus on why mirror life seems overrated as an existential threat. I know Jeremy because he is a recent high school graduate who reached out to us about volunteer opportunities at Apex Neuroscience and is now working with us on a review paper about ischemic brain perfusion. I predict that he is going to be a superstar one day.

17: New paper on progress towards the very long-term goal of developing molecular nanotechnology. And an interview between Max More and one of the authors, Ralph Merkle.

It’s interesting to see Ralph’s critique of Ken Hayworth’s traceability criteria for evaluating brain preservation protocols. I would love to see a debate between Ralph and Ken, perhaps someone could put that together. I think Ken is not appreciating Ralph’s point about the limitations of current electron microscopy vs future brain mapping technology. On the other hand, I think Ralph is not appreciating Ken’s point that it seems prudent to look at the structure of preserved brains and see whether it looks like the substrate that we think encodes memory is most likely still present. For those reasons I personally fall between them, as an Enlightened Centrist.

18: Our paper on methods for brain extraction in brain banking has now been published:

The complexity of cellular reactions is at once both daunting and awe-inspiring. And yet, now that we have LLMs, can’t we do anything? So maybe we could figure out how to simulate all of the chemical reactions in a cell.

There has been a lot of work on this over the years. For example, in 1999, a Japanese group introduced a software program called E-CELL. They used it to construct a model for a hypothetical cell with 127 genes (most of which were bacterial) that were supposed to be sufficient for transcription, translation, energy production, and phospholipid synthesis:

Progress continues. In March of this year, a new paper reported a simulation of the bacterium JCVI-syn3A. This included the transcription of all of its 493 genes, as well as metabolism, growth (based on the insertion of lipid and membrane proteins), and cell division, across a 100 minute cell cycle:

Zuckerberg and Chan’s nonprofit Biohub is also reportedly planning to spend 500 million on improving virtual models of cells.

Seems like a good idea to me. But it still seems like a really hard problem to use these models to actually design new drugs. First, the molecular reaction rates are often unidentifiable, which means that there are a number of ways that the parameter fits could work. Second, the molecular models are pretty high level, meaning that they do not actually have very much spatial or molecular-level detail. Third, accurately predicting how long a drug stays bound to even one molecular target is still fairly challenging. There are other difficulties too.

Neuroscientists have a tendency to annoy cell biologists by simplifying and abstracting away complex aspects of cellular biology when discussing the brain. In this post, I will continue in that noble tradition by cavalierly asserting that when it comes to simulating aspects of the function of the nervous system (perhaps many or even all of the cognitive ones people tend to care about), we can basically ignore most of those low level metabolic interactions in a given cell.

Specifically, I was recently listening to a talk by Bobby Kasthuri, co-founder of Meridian Neuro, when he made the point that it is easier to simulate a nervous system than a cell1. I liked this point because I found it on the one hand somewhat provocative, but then kind of obviously correct.

For example, in one model of the brain published in 2013, the molecular details that are important for neuronal computation can be boiled down to a short list. Cell size, ion channel type (e.g. Na or K channels), ion channel kinetics, and ion channel densities, together with the statistics of the stimulus under consideration, set the two things that control how a neuron transmits information, which are its speed and its signal-to-noise ratio.

We could have had (and in fact, we did have) this same conversation fifteen years ago, back when the only complete connectome belonged to a 302 neuron worm whose neurons do not fire conventional action potentials. The skeptics had a fair point there, as a wiring diagram (a somewhat incomplete one, but still) was mapped for decades without yielding much in terms of simulation.

But in the past few years, now that we have mapped the connectome of an organism with spiking neurons (i.e., that of Drosophila), we have empirical data showing that this connectome can be simulated in such a way that simple behaviors can be reproduced. This has been discussed to death already, so I won’t go into depth here, except to say that simple leaky integrate-and-fire models built on the fly connectome have made a variety of accurate predictions about neural activity and behavior. These results seems to have convinced a lot of people that simulating the function of a nervous system from its wiring is actually possible, at least for some behaviors.

However, it’s still unclear how far this straightforward simulation approach will go. The fly simulation studies focused on circuits with largely feedforward connections, for which the exact neuronal firing parameters may not really matter. Whereas simulating circuits with more recurrent connections might require more molecular data and modeling of non-spiking phenomena, such as neuromodulation.

Another counter is that maybe some or all of the cellular complexity is somehow required for neural computation, in which case simulating cells will be a prerequisite for simulating nervous systems (which is not impossible, as nervous systems are in fact made up of cells).

Ultimately, the proof is in the pudding. We will see in the coming years and decades the extent to which simulating a nervous system from its wiring ends up being a matter of inferring and filling in a few missing parameters or of reconstructing and modeling most of the underlying cell biology. It will probably depend on the circuit and function of interest.

Here’s an example of what they mean by the contactome. A particular part of a single dendritic spine has four other neurons that it shares synapses with (colored blue), while it has fifteen other neurons that its membrane contacts without a synapse (colored red, and seen individually on the right):

And here is an example of three neuronal somas (teal, purple, and red) whose membranes make a lot of contact with one another but don’t share any synapses — possibly they have some sort of other interaction:

They also found that astrocytes form a large syncytium-like network that spreads across the brain. This is well known from smaller scale studies — there’s a reason for the concept of a tripartite synapse — but this seems to be (one of) the first to describe it on such a large scale, with a single network containing 6146 of the 7562 total astrocytes in the volume.

They also find that the majority of synapses are <500 nm from a glial cell:

All in all it’s a nice study with some good visualizations. I do have some areas where it might be improved. First, it’s not always clear to me when they’re talking about glia broadly versus astrocytes specifically. This matters because astrocytes, oligodendrocytes, OPCs, and microglia all have very different functional roles in the brain in general, and viz a viz synapses in particular. I actually think that people should try to avoid using the term “glia”.

Second, what I would really like to know is how these cell-cell contacts might be functioning. They suggest a few possibilities in the intro: (a) gap junctions, (b) hormones, (c) neuropeptides, and (d) ephaptic coupling. It’s not entirely clear to me that these connectomes are at a high enough resolution to even visualize gap junctions, so I would have like to see that discussed. Ephaptic coupling is the most interesting possibility to me. It would be nice in a future study to model if that might be playing a role or not.

Finally, I’m not sure it’s fair to today’s field of connectomics to say it just maps synapses. I would say that’s more like “synaptomics.” Modern connectomics is usually more than that, including cell shapes and inferred cell types. Although I’m certainly not an expert in the field, I have previously defined a few different uses/levels of the term “connectome.” What I have called the “cell membrane connectome” is the 3D map of all cell membrane locations, which would include the contactome. So at least based on how I think about it, the contactome isn’t really a separate thing from the connectome, so much as a particular aspect of it that the field hasn’t emphasized much. And in fact, this paper uses the segmentations that the original “connectomics” studies already generated to extract its contactome.

That said, I do agree that these cell-cell contacts have not been studied or considered as much as they should be (probably because the field is just not very big), and maybe branding them with an -ome is a way to move the needle in that direction.

2: One of the world’s leading connectomics experts, Moritz Helmstaedter, has written an article about accelerating human brain connectomics. He argues that a human connectome could be completed within 10 years if $10-20 billion were invested.

He proposes that this be done in the context of a rapid body donation program. His threshold for high-quality ultrastructure is a 4 hour post-mortem interval (PMI), although he notes that even within that time window, connectome reconstruction becomes increasingly more difficult.

He notes that due to the PMI, two main problems develop: (a) synaptic vesicles are depleted, making synapses more difficult to detect, and (b) the very thin parts of axons (which are frequent in the grey matter) get “severed,” leaving only “spotty axonal traces,” and making them not possible to trace. Dendrite tracing can also be a problem at their thin spine necks, but generally dendrites are not as hard to trace because they are 5-10 fold wider than thin axons.

He also points out that it is usually possible even in highly damaged tissue to find some areas with relatively good ultrastructure for study, but that for a whole brain connectome project, the problem is much harder because it is the worst-preserved areas that are the bottleneck for success.

He notes that this need to lower the PMI raises “serious ethical concerns,” because for short PMIs, the exact time point and definition of death matters, and for this reason he calls for a discussion of the relevant ethics.

He considers immersion fixation as one potential approach for brain-wide ultrastructural preservation. In this analysis, he seems to be assuming that 3 mm is a short enough distance of immersion for fixative to penetrate the brain tissue fast enough. He notes that the surface of the cerebral cortex of the brain is exposed to CSF within 3-4 mm at maximum, which is encouraging. However, he notes that other areas, such as the basal ganglia, are much further away. Therefore, he rejects immersion fixation as sufficient and claims that perfusion fixation of the brain will be necessary.

The rest of the paper assumes that the brain has somehow been fixed rapidly enough so that the entire brain-wide nanometer-scale tissue is well enough preserved, and describes how it could be imaged at the ultrastructural level.

The main challenge is in finding a parcellation strategy that still allows neurite tracing. His proposal is that instead of intentionally cutting the brain into small pieces, it is actually best for the tissue to become brittle and then crack haphazardly into small pieces. These pieces can then be computationally reconstructed back into the whole brain, even if their order is lost, because they will each be so distinctive. Sort of like a giant jigsaw puzzle. Finally, the main cost is actually imaging them. He notes that 2,000 multibeam scanning electron microscopes working in parallel would allow the tissue to be imaged in 5 years, at a cost of $10-20 billion, which is the primary cost of the project.

Probably my main concern with the proposal is the idea to only store the synaptome (aka adjacency connectome). As I mentioned in the discussion of the “contactome” paper above, I just don’t think that the synaptome alone is likely to be that useful, and with all of the effort to image this brain, it seems like we should be able to do better at the data storage problem. It seems to me like some sort of mesh representation would not necessarily require storing the raw images but would allow much more information storage. There is precedent here, for example CERN can store an exabyte of data already.

It’s certainly a provocative proposal and I would love it if some government or other large funder got interested in it. I also completely agree with him that human connectomics would significantly advance the study of human brain disorders, indeed being very useful for understanding the cellular-level mechanisms of certain types of psychiatric disorders.

See also: Pelagia Martin’s article “The Connectome is Too Big for One Lab.”

3: A new computational approach for bootstrapping 2d annotations to 3d reconstructions of ultrastructural data. Includes this nice figure showing how the method can work with different input data types:

4: A software method for figuring out how to sample tissue banked brain tissue:

5: How does the precise organization of neuron wiring come about? A new study corroborates that in Drosophila, particular axons express unique RNA and protein molecules. This includes the cell surface receptor encoded by the Dscam gene, which undergoes alternative splicing and allows each neuron to express a particular combination of isoforms from tens of thousands of possibilities. Vertebrates have a functional analog of this that also allows for neurite self/non-self recognition (protocadherins).

Here are the axons they focused on (those of the neurons named “pSc” and “Gr59d”):

In addition to Dscam, this study also used RNA sequencing and found that Drosophila have a bunch of other transcripts for cell surface and secreted proteins that are differentially expressed between these two axon types (with multiple biological replicates of each):

The main reason I am interested in this is because it shows how molecular mapping could potentially be used by future technology to help decode the information in the original wiring diagram in the presence of damage.

6: Simulation of a cortical circuit based on connectomics data, cell types, and known synaptic physiology:

7: In a new study, living brains were profiled via two-photon calcium imaging of their cellular activity, then the tissue was fixed and sectioned for spatial transcriptomics. This allowed them to measure the RNA content of the same individual cells whose activity had been recorded in the live animal, with high spatial resolution.

8: A few years ago I looked into rapamycin for the purpose of trying to live longer. It was the hottest drug in longevity and, let’s be real, sometimes I follow fads. After some research, I sort of decided that aside from questions of efficacy, I was too concerned about infections, and I wasn’t sure whether I would also need (or be able to get) an antibiotic prescription as well. It seemed like some people prescribed them and some didn’t. I also felt bad about the idea of potentially contributing to antibiotic resistance without an actual medical indication (like in transplant patients who receive prophylactic antibiotics when they take it at a higher dose). It made the calculus worse for me. So I just never really pursued it, even though I always sort of had it in the back of my mind as something I “probably should have been doing.” Y’know, the whole death problem thing. But now some clinical trial evidence suggests that my subconscious was actually onto something!

Specifically, we got results from a randomized trial of adults aged 65-85 (n = 20 placebo, n = 20 of 6 mg rapa) for 13 weeks alongside exercise, with outcomes mostly related to exercise capacity. The primary outcome had no significant difference, but it leaned towards placebo (p = 0.089) and some of the prespecified sensitivity analyses significantly favored the placebo group. Granted this study doesn’t really measure what people tend to care about most with rapamycin, like the potential to slow aging or prevent some cases of cancer, but it’s not good if it worsens exercise capacity.

More troublingly in my book, the safety burden also seemed to favor placebo. There was one case of community-acquired pneumonia after a single dose of rapamycin, after which that participant was hospitalized and withdrew. More generally, adverse events attributed as potentially related to the drug were more common in the rapamycin group (35% vs 15%) and included upper respiratory tract symptoms, headache, and fatigue. Anyway, this data point seems to corroborate my priors for now, although I’m always open to changing my mind again in the future.

9: Mid-life elevated BMI is generally associated with an increased risk of Alzheimer’s disease. However, the “obesity paradox” in Alzheimer’s disease is the finding that late-life elevated BMI is often associated with lower rates of Alzheimer’s disease.

This is generally thought to be due to prodromal effects of Alzheimer’s disease leading to lower weight, but there hasn’t been all that much research on it.

A new study re-analyzes data from old Alzheimer’s trials with n = 1,663 participants to build a model to address this. They found that at lower amyloid levels (using the 25th percentile of the model for visualization), lower BMI was associated with better cognition (PACC) scores at baseline and over time. But at higher amyloid levels (e.g. at the 75th percentile), lower BMI was associated with slightly better cognition scores at baseline but a much faster decrease over time.

So the data seems to suggest that obesity really might be protective in adults with high amyloid, perhaps because of something like leptin signaling being beneficial in some way. However, I find their mechanistic explanations uncompelling and I still think it is most likely residual reverse causation.

This is because I think amyloid is not the whole story. Instead, amyloid is an imperfect proxy for disease stage. My explanation is that within the high-amyloid group, the people with lower BMI are more likely to be further along on the non-amyloid axes of disease burden (e.g. tau, neurodegeneration, hypothalamic dysfunction), which leads to lower appetite and feeding behavior, who would then be expected to have faster decline.

Whereas if people have low amyloid at baseline then they’re not having the amyloid cluster of problems in their brain, and lower BMI seems to be somewhat protective against cognitive decline for normal reasons like vascular function etc.

10: Obituary for Edna Foa, who helped establish exposure and response prevention (ERP) as the gold-standard therapy for OCD.

11: Study of an Italian cohort of n = 300 people with OCD finds that 18.6% (n = 56) were also diagnosed with obsessive compulsive personality disorder (OCPD). The group with comorbid OCPD was found to have a longer duration of untreated illness, of 9.2 ± 12.4 years vs. 6.1 ± 8.1 years.

12: A nice article with skepticism that whole brain emulation will matter for the AI transition.

I agree with a lot of it, but here are some disagreements because those are more interesting. The article seems to generally assume (a) rapid timelines, (b) somewhat rapid takeoff (to the extent that AI capabilities cannot be used to accelerate progress very much in the “wet lab” realm needed for emulation) and (c) no partial emulations.

The possibility of partial emulations is in my opinion the aspect that is most underappreciated here. An example would be only emulating the circuits for one or more particular neural functions, and simulating other inputs and outputs as needed. This might be useful for certain circuits that we value a lot, like those related to prosocial behavior, if that is ever possible.

Another example would be a broad category of “functional” emulations that capture higher-level properties of brain operations (such as those visible with neuroimaging) and use scale separation to abstract away from the underlying substrate. (On an even higher level, arguably this is what LLMs are already doing as simulators, and arguably there is a spectrum between emulation and simulation.)

Partial emulations would not qualify as identity-preserving for many people (even less so than full emulations!), but they could still be useful for other things, such as making a given AI system work in a more human-like manner.

I’m obviously not sure whether some sort of partial emulations will actually be relevant in the medium term, but I think it’s more complicated than the binary “full richly molecular annotated nanoscale-level human WBE or bust because of AI timelines” story.

I also predict eventually people will have debates about whether emulation has happened yet, in the same way that people currently can debate whether we have already achieved AGI.

13: Nice discussion by Nikki Olson on long-term monitoring systems for detecting cardiac arrest for the purpose of biostasis.

14: Our article “Physician estimates of the feasibility of preserving the dying for future revival,” has now been published. PLOS kindly did a press release for it, which is available here. Here’s one of the main figures, perhaps sort of self-explanatory:

Much has been written about this idea over the years. A group of neuroscientists has now written a paper that takes it more literally than most. Specifically, they propose to build an “ultrastructure-to-dynamics compiler” for the brain.

In computer science, a compiler is what translates a program written in a human-understandable language into something that a machine can actually execute. The “neural compiler” proposed in this paper would take detailed images of brain structure, possibly including some degree of molecular annotation, and then translate that into the parameters needed to simulate how the circuit behaves.

Their reasoning is that on one hand, anatomical imaging is getting cheaper at a relatively fast pace (although I certainly wish it were faster and cheaper and better!).

But on the other hand, physiological recording is improving relatively more slowly. The compiler idea is meant to address this problem by using paired structure-function data sets to learn how to predict dynamic parameters (e.g. the synaptic conductance) from morphomolecular data alone.

They propose to test the idea in the retina, which is highly characterized and has well defined inputs (photons) and outputs (neuronal spikes).

They note that it would not necessarily be possible with current technology to predict the dynamics with arbitrarily high accuracy, at least if there is little or no molecular annotation. Instead, they propose that the compiler should constrain the space of plausible models, with the degree of constraint possible depending on the degree of molecular annotation. The output would be a probability distribution over possible parameters, which could then be further narrowed once it is modeled as a part of a larger network with its own constraints.

The article reads sort of like a grant proposal. But it’s a grant that I would like to fund if I had the capacity to do so! It’s clearly an important problem where more systematic data and analysis would be helpful.

2: One of the long-standing questions people have raised with aldehyde fixation is the extent to which it affects molecular properties of the tissue.

A new study helps to address this question. First, they compared the proteome (i.e. all of the measurable proteins) from fresh-frozen compared to formaldehyde fixed mouse tissue. They found a high correlation, with no evidence of extraction bias affecting any particular cellular compartment.

Next, they developed a method for laser capture microdissection to capture the proteome of single neurons from fixed mouse brain tissue.

They found that they could recover around 1,500 protein groups per single neuron on average, and were able to recover more proteins when they sampled larger areas. The proteins that were missed in single cell samples were mainly low abundance ones. This is consistent with the limits of mass spectrometry in general, as opposed to the loss of a particular protein type due to fixation.

Overall, I consider this to be another solid data point that high-quality fixation preserves the molecular constituents of the brain in their original locations.

3: A new study characterizes a human brain that was immersion fixed 7 hours after death with multiple modalities, including MRI, synchrotron imaging, micro-CT, and electron microscopy.

The goal of it is to be a prototype study to show that they can map white matter architecture in human brains across scales (a buzz word, but accurate here). They showed that it is possible to see single myelinated axons using non-invasive synchrotron imaging as well as micro-CT.

Here is some of the synchrotron imaging data they collected (i.e. Hierarchical Phase-Contrast Tomography, or HiP-CT), which shows individual axon fascicles:

In their electron microscopy data, they are able to see individual myelinated axons. However, they note that there are some regions with de-compacted myelin (red arrows).

In the discussion, they note that specimens with shorter postmortem intervals would provide better ultrastructural preservation. It’s interesting but not particularly surprising to me that they were able to achieve this level of preservation quality via immersion fixation 7 hours after death.

4: Another methods paper for studying white matter connections in postmortem human brain tissue, this time using 3x hydrogel expansion, immunolabeling for neurofilament H, and light-sheet microscopy imaging. See also the companion methods paper.

The brain tissue was from a 61 year old male donor obtained from the San Diego Medical Examiner’s Office. The brain was removed and bisected at the midline, and coronal slabs approximately 0.5 cm thick were cut. These were embedded in alginate, which provided mechanical support, then rapidly frozen in a dry ice-isopentane slurry, vacuum sealed, and stored at -80°C. When retrieved from the -80°C freezer for processing, the frozen slab was immersion fixed in ice-cold 4% paraformaldehyde with 10% acrylamide in PBS for 24 hours at 4°C. After all of this, they were still able to see individual axons stained with neurofilament H:

They also provide Neuroglancer links in which you can view the raw data yourself:

A key open question to me is regarding the size of axons they are able to visualize (and thus demonstrate the preservation of).

The abstract notes that they can see trajectories of densely stained large projection axons greater than ~1 μm in diameter. In the results, they note that there was generally a good amount of spacing between axons and that those with a larger caliber, above ~1.5 μm, could be easily traced manually with few ambiguities.

Intriguingly, they wrote in the methods: “NFH was selected based on its established use for myelinated long-range projection axons. Control experiments with viral labeling of excitatory neurons in macaques demonstrated that all virally labeled cortical axons were also labeled with NF antibodies (data not shown [can be included in revision])”.

One thing I am interested in is what fraction of axons are actually being visualized. Their ~1 μm and ~1.5 μm size thresholds are for what can be clearly resolved and easily manually traced, but it seems that they cannot address how many axons fall below those thresholds. 

The main reason I care is that I want to use their results as a proxy for what kind of axon preservation quality we can expect in human brains subjected to the relatively harsh conditions reported here, including non-cryoprotected (but rapid) freezing, immersion fixation after thawing, and presumably a decently long postmortem interval (although they do not report it).

It’s confusing to me that all of the large axons look so good here, and yet axons in general often look damaged in other data sets (like electron microscopy) that have undergone similar types of preservation steps. Why is that? Is it just that the smaller axons that are seen more commonly on electron microscopy are more vulnerable to damage? But why are they so much more vulnerable? And if thin axons really are just more fragile, does the good preservation of the large axons suggest that the way smaller axons break down is constrained enough to be likely to be inferable from the breakdown products? Or is there something else going on?

I realize this is probably inside baseball to 99% of my readers, and I apologize about that, but it’s something I’m thinking a lot about right now, so perhaps you might find it interesting.

5: Study finds a similar distribution of phosphorylated tau in people with Down syndrome with Alzheimer’s disease neuropathology (DSAD) as people with late-onset Alzheimer’s disease (LOAD). This is despite people with DSAD being substantially younger on average.

6: Study finds that dementia in people with schizophrenia seems to be a distinct entity, insofar as its pattern of demographic, clinical, and genetic correlates cannot be easily explained by comorbid Alzheimer’s disease or cardiovascular risk factors.

7: A randomized trial finds that more hydration doesn’t actually help prevent kidney stones, but does cause more nocturia. Another L for Big Water.

8: Two randomized trials (n = 3808 participants total) find that oral semaglutide is not effective at slowing the progression of disease in early Alzheimer’s. So another drug enters the Alzheimer’s disease randomized trial graveyard.

Hindsight is 20/20, but this is not super surprising to me. People with Alzheimer’s disease often have lower BMIs to begin with, for reasons unclear to me.

One interesting note is that there were 5 fatalities attributed by the (blinded) investigators to the study drug in the trial, but 4/5 of them were in the placebo group, which was half of the participants! This is a nice example of how it is hard to figure out what caused someone’s death. And also (weak) evidence of how relatively safe semaglutide is from a mortality perspective.

9: Is social media time actually bad for youth mental health? A meta-analysis of 46 studies finds that the “current pool of research is unable to support claims of harmful effects for social media use on youth internalizing disorders.”

It’s an interesting abstract and may be correct. However, I take issue with their use of the term “there is no evidence that” in the title. People should stop using that phrase once and for all. There is nearly always evidence of things on both sides of any active debate. The question is nearly always about what the preponderance of evidence best supports. I know this point has been made many times before, but apparently it has not yet reached everyone’s cortex.

10: Kevin Kennedy has an article with some nice criticism of the SAINT trial of TMS for depression.

11: A case of HPPD reported in a clinical trial of 25 mg psilocybin for treatment resistant depression. “One participant with a history of anxiety and dissociative symptoms acutely experienced severe panic, which subsided first, but reoccurred in form of anxiety and panic attacks, dissociation/depersonalization, flashback phenomena and symptoms of hallucinogen persisting perception disorder (HPPD) requiring hospitalization.” The overall trial did not show a significant effect on the primary outcome.

12: An article on the HHMI approach to journal publications. They are trying to move away from journal publications and towards evaluating preprints. In order to do this, journal names are now stripped away from the citations in the HHMI review submission materials. Also, at HHMI’s Janelia campus, scientists are not allowed to write the journal name on their presentations.

This is such. A. Good. Idea.

Chasing journal prestige may be good for individual scientist careers (actually, I’d bet that on average, it is), but it is a collective action problem that ends up enriching publishing companies at the cost of precious taxpayer and philanthropic dollars for research. Incentive realignment to prevent researchers from chasing prestige over actually advancing science is exactly the way to solve this problem.

This is why, at NN, I try to avoid mentioning the journal name of articles, and I try to ignore this when I choose which articles to write about.

13: The Amaranth Foundation on why they are funding neuroscience research in order to get safer AI, as a form of defensive accelerationism.

14: Interview with the founder of Bexorg, a new neuroscience company that is using perfusion of postmortem human brains as a platform to investigate new drugs. I am thrilled that they were able to raise 42.5 million dollars for this, because maybe the market for this type of human neuropharmacologic research is larger than I had realized.

15: Mark Woodward’s custom liquid handler at his cryobiology company Wake Bio.

16: There is a literature on using focused-beam microwave irradiation to rapidly heat rodent brains to 75-90°C. Here is a recent paper I just saw on this, which reminded me of it.

Microwave irradiation can denature enzymes and halts post-mortem metabolism within a second. The argument is that most of postmortem decomposition is enzyme-mediated. So if you could rapidly inactivate enzymes, you would dramatically slow autolysis, and so you would have much more time to preserve the structure of the brain.

I’ve looked into applying this to human brains, but the main problem is penetration depth. Microwave energy at these frequencies only penetrates a few centimeters into the brain tissue, which works for a mouse brain but wouldn’t work for a human brain.

17: A paper from last year on the pathophysiology of the agonal phase, which is a highly understudied topic very relevant to brain preservation (HT: Mike Darwin).

Specifically, they studied 39 critically ill patients undergoing withdrawal of life-sustaining treatment while monitoring their cerebrovascular physiology until circulatory arrest occurred.

They found that in general, cerebral blood velocity declined linearly with hypotension. However, brain blood flow velocities and brain tissue oxygen tension stopped before systemic circulatory arrest. The median time from the cessation of brain perfusion to pulseless electrical activity was around 1.5 minutes, and the time to electrical asystole (ECG flatline) was around 6.5 minutes.

Because legal death is often declared a couple of minutes after pulseless electrical activity, the “warm ischemia” clock for the brain probably starts several minutes earlier than the time of declaration suggests.

18: Review by Brian Wowk on rewarming cryopreserved tissue with electric fields.

19: Study on the attitudes to cryopreservation among people living in Switzerland. In general the public is highly supportive of the practice being legal, with 83.6% of respondents expressing no support for legal restrictions nor making a change from its current status. However, that is a bit of a low bar IMO. Surprisingly, 20.1% expressed interest or intent to sign up for preservation services.

20: Very interesting interview of the superstar brain preservation researcher Alex German by the journalist Kristen French.

Q: Would you personally want to be cryopreserved?

A: If the alternative were facing certain death, I currently would. It’s important to note that you expose the person to a lot of uncertainty in the future. They become more vulnerable to long-term variance in the developmental trajectory of humanity. There might be risks that we’re not aware of today. In a representative survey of 1,000 Germans, about 22 percent of them said they want to undergo these procedures, the majority of these being male.

My personal preferences could change in the future. But my professional opinion, as a physician, is that patients with bleak prognosis, and the public in general, should be informed about these possibilities. And these should be discussed in a measured, rational way. And this is what we try to contribute to with our research.

The two major neuron types are excitatory and inhibitory. Most of the neurons (around 80-90%) in the cerebral cortex are excitatory (meaning that their synaptic transmission makes downstream neurons more likely to fire an action potential), while around 10-20% of them are inhibitory.

It turns out that despite being a minority, inhibitory neurons are much more diverse than excitatory neurons (in terms of anatomy, gene expression, and electrophysiology), which naturally makes them more interesting.

A new study examines inhibitory neuron typing in the whisker barrel cortex of the rat:

The authors note in the introduction that there is no consensus on how many inhibitory neuron “types” there are. However, this is utterly unsurprising because various aspects of neuronal features (including electrophysiology) are on a multidimensional set of hierarchical spectrums, so there’s generally no objective answer to how many types of a neuron there are.

The authors recorded electrophysiology from individual cells, preserved their brains via perfusion fixation, and then labeled and imaged the cells via light microscopy:

They first clustered the neurons that they profiled based on both their morphology (B) and electrophysiology (C):

They then clustered neurons into a space that combines both morphology and electrophysiology, which they call “morphoelectric”. They noticed that these clusters are highly correlated with the depth of the cell body in the cortex (from layers 1 to 6):

They found that the morphoelectric properties of a neuron could predict which of the classic four molecular markers of inhibitory neurons it expresses (PV, Sst, Vip or Lamp5 ). The accuracy was pretty high:

Orthogonal to a given cell’s molecular expression, they also found that cortical depth was predictive of the morphoelectric properties of cells, in different data sets:

The authors speculate that the cortical depth has such a large effect independently of the molecular cell type because it is associated with local environmental effects on the morphology and electrophysiology of cells.

Extrinsic environmental effects on neuronal function are a potential problem for the connectome + cell type = simulation idea. Some extrinsic environmental effects, like the ones described in this paper, might be predictable based on the depth of the soma in the cortex, in which case you could extend the equation to connectome + cell type/cell location = simulation.

But other extrinsic environmental effects on neuronal function are not necessarily going to be easily predictable based on the location of the soma. They might require inferring the functioning of local glial cells based on morphological images, or doing molecular profiling. The extent to which this will matter for simulating any particular type of neural function is very much an open question.

2: Speaking of brain simulation, the nascent brain emulation company Eon Systems went a bit viral this month with an announcement from their founder Michael Andregg that “We’ve uploaded a fruit fly.” More detail was eventually described in the Eon Systems blog post.

Ken Hayworth expressed skepticism. So did Michał Januszewski, who noted “extraordinary claims require extraordinary evidence.” Alexander Bates had some good questions:

I actually think Bates’s grandmother may have the best point of all: “But - I thought flies didn’t have brains?” (from Bates’s personal website)

3: Speaking of the head direction ring attractor bump, what exactly is that, anyway? And what kind of research are people doing on it right now?

First, a bunch of background. The term “manifold” has been one of the hottest in neuroscience over the past decade:

When neuroscientists use the term “manifold”, they use it differently from the formal mathematical definition.

Instead, they are generally referring to the common phenomenon that neural activity, which technically exists in a high-dimensional space (i.e. one dimension per neuron), usually actually “lives in” a lower-dimensional surface, called a manifold. A classic example is a ring, which is a one dimensional manifold embedded in a two dimensional space. Here’s a toy example of a ring-like manifold of neural activity in a higher dimension space:

Another important concept here is an attractor. In a dynamical system, an attractor is a state that the system tends to settle into over time.

A ring attractor, then, is a neural activity space that is constrained to a ring manifold with attractors at different discrete points. An energy landscape is one way to visualize this. Low points (valleys) are the states the network naturally settles into, while high points are the states it tends to move away from.

Neural state spaces made of large numbers of neurons tend to have smooth energy landscapes (top), while spaces made of small number of neurons (bottom, n = 6 here) tend to have “bumpy” energy landscapes:

Computational models of neurons that have local excitation and diffuse inhibition can create a ring-like attractor system. And as it turns out, this is a very good way to model how flies maintain their head direction when moving through space:

With the recent mapping of the fly connectome, we now know how the neurons responsible for maintaining head direction connect to one another. These are called “compass” neurons. You can see the local connectome and also an example of a calcium imaging activity map:

Bates’s point — which is a good one — is that anybody who claims to have “uploaded” a fly should be able to tell that these internal dynamics are working as expected. Which is challenging, because we don’t even know exactly how they work yet!

Finally I can get to a recent paper which can be used to show some of the uncertainty in this field.

A key structure that helps to maintain head direction is the ellipsoid body, which is a donut-shaped structure made entirely of axons and dendrites. The cell bodies of the compass neurons whose neurites project into it are found elsewhere in the fly brain.

The ellipsoid body can be divided into 8 wedges. Each wedge has processes from a different group of compass neurons. Here is a reconstruction of the ellipsoid body connectome, color-coded by wedge:

At any given moment, the compass neurons in a few adjacent wedges are more active than the rest, forming a localized “bump” of activity. As the fly turns, this currently-active bump travels around the ring, so that its position at any given moment encodes the fly’s current head direction.

This study also uses calcium imaging (electrophysiology) data from the ellipsoid body, where darker regions indicate higher neural activity, an example of which is here:

What this article found was that the fly connectome is well positioned in the space of possible connectomes for supporting robust ring attractor dynamics. Even when synapse counts were varied by nearly 90%, the scaling parameters could nearly always be adjusted to restore ring attractor functionality.

In particular, even though all four of the available fly connectomes (male CNS, hemibrain, FAFB, and BANC) have different connectivity in this area, they all fall well within the region of connectome space where ring attractor dynamics are achievable. This means that evolution seems to have found a connectivity pattern that is tolerant of natural synaptic variability between flies:

However, we do not know what the scaling factors that the authors used to tune the network actually correspond to biologically. The authors suggest they could reflect neuromodulation, which can up- or down-regulate synaptic strengths across local groups of cell types all at once.

Could these scaling factors be inferred purely from electron microscopy data, given enough “side experiments” to figure out how they work? Hard to say, but perhaps. Either way, we definitely don’t have this kind of information yet, which is another reminder of how far we are from being able to simulate even relatively simple neural circuits at a very detailed level.

4: A new study on the human brain thanatotranscriptome, i.e. gene expression changes occurring after death. They find that postmortem time induces artifactual upregulation of genes involved in oxidative phosphorylation, ATP synthesis, and inflammatory responses. Notably, 6 hours of storage of the brain at 4°C (sort of mimicking rapid morgue refrigeration) produced zero genes that were differentially expressed by their significance thresholds, which suggests that the changes in gene expression after death are not necessarily immediate and huge.

5: A big retrospective study of n = 606,434 US veterans with type 2 diabetes finds that, compared to SGLT-2 inhibitors, GLP-1 receptor agonists were associated with an around 13-25% relative risk reduction of developing new substance use disorders (i.e. alcohol, cannabis, cocaine, nicotine, and opioid use disorders) over three years of treatment.

6: Nils Wendel on the new one year open-label study of the non-dopaminergic antipsychotic xanomeline. It continues to have a low side effect burden, including no significant EPS or metabolic effects, which is obviously just amazing.

7: A nice essay about schizophrenia trajectories by Michael Halassa. He covers a GWAS-by-subtraction study which suggests that there may be two components within the schizophrenia GWAS signal: (a) one component associated with lower educational attainment, with mechanisms more related to early neurodevelopmental genes expressed across the brain, and (b) one component associated with higher educational attainment, with mechanisms more related to synaptic function later in life that tend to be found more in the frontal cortex. It’s a preliminary hypothesis that requires more study, but I find it interesting.

8: A study on the effect of long formalin fixation periods on the immunohistochemistry of different vascular markers in the postmortem human brain. Claudin-5 is one antigen that was found to not really change, even after 20 years of storage:

9: Study of archival brain tissue stored in formalin fixative at room temperature for decades. They reported: “In general, good staining quality was achieved across all decades, ranging from 1946 to 1980, and across both types of tissue preservation, i.e. original blocks stored in cardboard boxes for up to 78 years and newly sampled blocks with fixation times up to 78 years.”

10: A new paper from Adam Higgins’s lab describes a robotic platform to identify the minimum concentration of cryoprotectant needed to vitrify to find lower toxicity cryoprotectant formulations:

11: Alexander German et al’s article describing the use of interleaved equilibration to maintain electrophysiological properties of brain tissue after cryopreservation is now published in a peer-reviewed journal. This has been widely covered in the scientific press (for good reason), but loyal NN readers already knew about it last February when the preprint was published.

12: A new paper from Aurelia Song et al on the use of aldehyde-stabilized cryopreservation to preserve the ultrastructure of whole pig brains. They report unsuccessful preservation in one case where the cannulation of the aorta was not successful (A, C, E) and successful preservation in one case where perfusion started 14 minutes after cardiac arrest (B, D, F):

In the successful preservation case, they also report electron microscopy data, one image of which can be seen here:

While these are amazing and encouraging results, I’m not sure if they will necessarily translate to humans in the context of medical aid in dying (MAID), which the authors suggest in the paper. There are many potential differences between these situations, including humans having a 10x larger brain, generally being much older and likely having some degree of vascular disease (e.g. atherosclerosis can begin as early as age 2), and having a longer agonal period (even with MAID). And plus, biological variability means that perfusion might work well in some cases and less well in others, even at the same time point.

13: On a related note, Aurelia Song and I discuss some of our differences in opinion about brain preservation in this Less Wrong comment thread.

What did they find? One finding is that most risk variation appears to predate the divergence of the African, European, and East Asian ancestral populations. The way they could tell this is because the effect sizes of genome-wide significant variants discovered in an independent European cohort were significantly correlated with the effect sizes of these variants in European, East Asian, and African samples considered separately. The vast majority of risk variants point in the same direction across all three of these groups.

I find this interesting evolutionarily because if the risk architecture is mostly ancient and shared, it raises the question of why selection over the eons has not eliminated these risk variants.

When it comes to using this genetic risk for prediction of schizophrenia risk in individuals, the odds ratio are stronger in those of European ancestry. The authors attribute this to the lower power of the available African ancestry training data.

What do their results tell us about the biology of schizophrenia? The picture they paint is still pretty coarse. Cell type enrichment analysis found a significant enrichment across basically all subtypes of excitatory and inhibitory neurons, a minor signal in oligodendrocyte precursor cells, and no significant signal in other non-neuronal cell types.

They also report that risk variants are enriched for genes in both presynaptic and postsynaptic compartments. But I am skeptical that this synaptic compartment enrichment tells us much. What would be more interesting is if the synaptic compartment of neurons were found to be differentially enriched compared to some other compartments of neurons, like the cell body or axons or something. I haven’t seen that.

It’s also notable to me that even though schizophrenia has such strong clinical evidence for immune involvement, there was no immune cell type signal enrichment. In fairness, the MHC region, which has previously been found to contain strong GWAS hits for schizophrenia, was excluded from the analysis due to its complex genetic structure. But the MHC region aside, when it comes to common genetic risk variants, something seems rotten in the immune hypothesis of schizophrenia.

2: Neuronal dendrites can be split into shafts and spines. Spines are small protrusions off of the main shaft. They are often the site where a synapse from a different neuron is located, especially for excitatory neurons. It’s also common that multiple synapses are present on a single spine, classically one excitatory input and one inhibitory one. Dendritic spines are a key part of neuroanatomy because they allow for the biochemical isolation of signals, and they grow and shrink with learning. Basically, you can’t care too much about dendritic spines. You are made of them! (Alongside, in fairness, a bunch of other stuff.)

It turns out that identifying spines with computers is not the easiest thing in the world. A new study develops a computational method to automatically identify spines, shafts, and somas in connectomics data sets.

They then use this to identify where synapses are located. For some additional background, neurons can be classified as excitatory (in which case all of their synapses increase the likelihood that the downstream neuron fires) or inhibitory (in which they all decrease it).

It turns out that excitatory and inhibitory neurons have very different patterns of dendritic spines. 70.3% of the synapses onto excitatory neurons are onto spines, but that number is only 19.8% for inhibitory neurons. Why is that? Inhibitory neurons simply have fewer spines on their dendrites, as most of their dendritic surface is shaft instead. And why is that? Well, the authors don’t really address that. It seems to be an open question.

3: Speaking of dendritic spines, a new computational modeling study evaluates the anatomy of around 2000 dendritic spines. They find that constrictions in the neck of the spine substantially increase the electrical resistance of this compartment, more than has previously been appreciated. This means that an inhibitory synapse at a dendritic spine can act as a “veto gate” that can rapidly turn off the signal from an excitatory synapse present there.

4: How do myelin sheaths “know” how large to grow? A new study finds that the protein TMEM63A acts as a mechanosensitive ion channel in oligodendrocytes. It senses the degree of membrane stretching during axon wrapping and converts this into calcium signals that guide the thickness and length of myelin sheaths to match the axon diameter.

5: A clinical trial from the Netherlands that came out a few months ago, the HAMLETT trial (n=347), randomized patients with first-episode psychosis after remission to early antipsychotic dose reduction/discontinuation (DRD) or maintenance. This is a very important type of study. They didn’t find very big differences:

You can see that participants in both groups ended up taking the same average dose of antipsychotic medicine. The primary outcome, patient-rated functioning, had no differences.

They found that the dose reduction/discontinuation group had significantly better outcomes in the secondary outcome of clinician-rated functioning (GAF), arguing that this resulted from a “learning experience” due to guided tapering. But it doesn’t seem that they corrected for multiple hypothesis testing and I highly doubt that this result would have survived that. And I find the learning experience idea slightly too cute.

3 people died of suicide in the dose reduction/discontinuation group vs. 1 in the maintenance group. The study wasn’t powered to detect mortality differences, but it’s notable.

Lex Wunderink was an author on the earlier studies that found long-term functional benefits of dose reduction/discontinuation and that motivated this trial. He notes that the intention-to-treat design means that once the doses converged between the arms after year 1, no strong conclusions can really be drawn about dose reduction/discontinuation vs. maintenance.

While most of my take aways here are a bit skeptical of the interpretations, I want to note that I really like this study and think that many more studies like it should be done.

6: Changes are coming to the next DSM. First, they’re proposing renaming it the “Diagnostic and Scientific Manual” (replacing “Statistical”). Their reasoning for this makes sense, but as a statistics guy, I can’t say I love it. Perhaps eventually it will just be DSM and not stand for anything, like the SAT.

They note that they want to embrace biology and environment as key determinants rather than remaining agnostic about etiology. This seems perilous, but possibly necessary. Because they note that as neuroscience elucidates the pathophysiology of more psychiatric conditions, those conditions risk being “removed” from psychiatry’s purview (as happened with Rett syndrome), potentially eroding the field. They want to avoid a dualism where only disorders with unknown biology count as psychiatric. I certainly agree with that.

7: A new systematic review on tachyphylaxis (i.e. where a drug’s effect diminishes rapidly within hours or days) and tolerance (i.e. a gradual loss of drug effectiveness over longer-term use) to medications used for ADHD. They found that tachyphylaxis occurs for the affective/behavioral effects of stimulants and that tolerance develops to subjective effects such as “drug liking” in neurotypical volunteers. However, they found little to no evidence that tolerance develops to the therapeutic or cardiovascular effects of ADHD medications in clinical settings over the longer term.

My guess is that the same is likely true of other stimulants, like caffeine.

8: A review article on the Japan Brain Bank Net. It includes 17 brain banks across Japan. They have a particular focus on psychiatric disorders, and many psychiatrists are now collaborating. I think this is fantastic because postmortem autopsy is really just an irreplaceable way to learn about the pathophysiology of disease.

9: Another health trend I got into in the 2010s bites the dust (like 10,000 steps a day), as Patrick Mineault explains why blue light filters don’t work for improving sleep. However, he notes it is probably helpful for sleep to go to dark mode or decrease screen brightness at night, and to get more light during the day.

10: Sasha Chapin’s communication tip: it always feels bad when someone you are talking to is trying to constrain your agency. Instead, it’s better to communicate with genuine openness about what sort of response might come back at you.

11: A new study of two human lung specimens preserved in spirits of wine (alcohol) for hundreds of years. They find that they are able to reconstruct a full RNA virus genome from these specimens, even without formalin fixation.

One question I’m interested in is what kind of neural information might still be recoverable from centuries-old alcohol-preserved brain specimens.

Of historical note, the brain of Charles Babbage (considered the “father of the computer”) was preserved in ethanol in 1871 after his death. Here are some photographs of one hemisphere of it from 1909:

12: An analysis of how long billionaires live relative to the population average. Not much, only 0-4 years more. Ends with an argument that they might want to contribute more to longevity research (I agree, but I’m a bit biased).

13: In 2019 Mati Roy created a list of all the people who have ever had their brains preserved with the goal of future revival. He found that it was doubling every 9 years. This has been called “Roy’s law.”

He has now updated the list. He found that it has increased from around 466 in 2019 to around 716 now. The doubling rate is still around 9 years. The iron law of straight lines on curves continues.

14: Mike Darwin with a nice list of why many people have not signed up for cryonics and are not interested in it.

15: Jordan Sparks with predictions on how robots might become relevant in the future of brain preservation.

16: An academic paper argues that individuals who sign cryopreservation agreements should be legally classified as “cryo-consumers” and therefore protected by consumer protection laws, while those who are already preserved should be classified as “cryo-patients.”

17: A palliative care team in Sarasota, Florida reports on the case of a 78-year-old cancer patient who had arranged for cryonics with an out-of-state company. After he lost capacity, the cryonics company’s requests for post-mortem chest compressions and heparin administration created ethical challenges for the medical team, who declined to perform either. They did perform cooling of the body on ice, and it was eventually transferred to the cryonics facility.

I have to wonder if cryonics became more popular if medical teams would become more comfortable with interventions like these.

As a result, in a successful cryoprotectant perfusion (without fixatives), water rushes out of the tissue into the blood vessels, the tissue dehydrates, and you end up with a shrunken brain that is visibly pulled away from the skull. The brain weight goes down by 50% or more. This is currently considered a good sign of cryoprotectant perfusion quality.

Far be it from me to say that a brain preservation method will not work because it seems weird. I myself have proposed that aldehyde fixation — something which is definitively lethal by contemporary medical criteria — may allow people to be revived with meaningful memories intact if humanity develops sufficiently advanced technology in the future. So I’m not going to use the absurdity heuristic here.

Instead, the key question is what this severe dehydration does to the nanometer-scale structures in the brain, such as the connections between neurons, that are thought to be the key parts of the information that encodes long-term memories.

Previous attempts at imaging this type of brain tissue were stymied because the severe dehydration made the tissue look unrecognizable. Synapses could be seen, but it wasn’t possible to clearly identify individual neurites or trace them to see whether the connectome is intact:

A new paper from Greg Fahy et al at 21st Century Medicine provides the most detailed look yet at what happens to brain ultrastructure during vitrification. So naturally, I had a look at it.

What do they think of non-vitrification approaches?

First, the paper opens with a very interesting review of prior work on brain cryopreservation, including some original data from the ever-controversial experiments of Isamu Suda performed in the 1960s.

The authors are not very enthused about freezing-based approaches. They show electron micrographs of rabbit brains that were perfused with glycerol and slowly frozen, which have large ice cavities that grossly distort the tissue:

Their position seems to be that vitrification — i.e., cooling without ice formation — is the only serious path forward for brain cryopreservation. But vitrification requires especially high concentrations of cryoprotectants, which causes the severe dehydration that has made it difficult to assess whether the connectome is preserved.

Rabbit experiments

They performed two types of rabbit experiments.

In the first group, rabbit brains were perfused with M22 for 60 minutes, vitrified, rewarmed, and then fixed with a solution that still contains the same concentration of cryoprotectant. This shows us what the tissue looks like in its shrunken state. At high magnification they can identify synapses, mitochondria, and what appears to be some morphologically intact cell membranes. But everything is compressed, making it difficult to clearly distinguish or trace individual cellular processes. These look similar to the previous images of vitrified brain tissue.

In the second group, rabbit brains were perfused with M22 and then gradually diluted back to a lower concentration of cryoprotectant before fixation — to 5M, 3M, or 1M. This tests the extent to which the shrinkage is reversible.

Although notably, it seems to me that most of the shrinkage is expected to occur between 0 and ~3M concentration of cryoprotectant. So my understanding is that diluting from full M22 back to 3M or 5M wouldn’t be expected to reverse most of the dehydration, although I might be wrong about that:

Anyway, they found that at 5M, the tissue is still quite shrunken. After reversal to 3M, though, things look considerably better, as the neurons have more normal-looking apical axons, and synapses with visible presynaptic vesicles can be identified. This is probably their best-looking EM data. However, there are various forms of damage, such as the wavy intracellular white spaces, that I don’t understand the cause of. Also, there are some places in the zoomed-in image (C) where I can’t really tell whether I am looking at two smaller processes or one larger one:

When they reverse to 1M (Fig 10), things go a bit wrong. While they can see nicely preserved synapses with clear presynaptic vesicles, they also see what they call “exploded” neurons, which they attribute to osmotic damage from removing the cryoprotectant too quickly.

They report that this problem is solvable, by using osmolytes to counterbalance the intracellular cryoprotectant during washout. They report that this can prevent this ultrastructural damage “even when all M22 is washed out of the brain.” But this data is not presented in the paper. It’s cited as “Spindler et al., in preparation.”

Human data

The human data comes from a single brain, that of a 73-year-old terminal cancer patient who donated biopsy samples of his brain for this research. His brain had significant ischemic injury before preservation even began, consisting of two days of agonal hypoxia before legal death, then three hours of cold ischemia before perfusion started. That’s important because it’s actually an example of realistic brain preservation conditions.

Cortical biopsies were taken after whole-brain M22 perfusion, plunged into liquid nitrogen, and stored for four years. They were then warmed and processed in different ways.

Some biopsies were warmed straight into fixative containing M22, showing us the fully dehydrated state. As expected, it is severely shrunken and electron-dense, but without obvious ice damage. On electron microscopy, they can identify synapses and some intact-appearing membranes.

Other biopsies of the human brain were rewarmed into diluted M22 (75% or 66%) before fixation. Using light microscopy, they report that this partial rehydration caused cells to regain their general characteristic shapes, with neurites visible.

However, the rehydrated human tissue was only examined by light microscopy, not electron microscopy.

One of the main concerns with vitrification has long been that the severe dehydration might be masking damage to the structure of the brain. For example, if the cell membranes are broken apart or there are areas of washed out structure due to ischemia and rapid osmotic shifts during perfusion, we might not be able to see it when everything is compressed together.

Because the electron microscopy that they did show from the human brain tissue is still so compacted, we can’t really evaluate for the presence or absence of such damage yet with much certainty.

Summary

In brain preservation, as much as possible, we need objective metrics. One metric that has been proposed by Ken Hayworth and Sebastian Seung is to see whether it is possible to trace the connectome of a preserved brain. This is widely viewed by researchers as one of the best available metrics of preservation quality.

In ideal laboratory animal settings (rabbits), the key result is that they show partial reversibility (to 3M) with improved electron microscopy quality, but still without enough reversibility to see clearly traceable processes across the 2d image. They report that complete washout is actually possible with a new method that they developed, which is fantastic news. But this rests on unpublished work that we will have to wait to see in the future.

Ideally, they would show in this future work that they can reliably trace the connectome across randomly sampled areas of the brain, which would allow the pure vitrification approach to reach parity with aldehyde-based methods in ideal laboratory animal settings, and shift the debate to which method is the best at structural preservation in realistic settings.

In the single human case, they show that perfusion-based vitrification is feasible in at least some parts of the brain even hours after legal death, and that cells regain their general characteristic shapes after partial rehydration. But the rehydrated human tissue was only examined by light microscopy, not electron microscopy, so we can’t tell whether the connectome is likely to be traceable in the local area where this biopsy sample was taken from.

This paper is clearly a step forward and a very important contribution to the brain preservation literature. I would like to personally thank the authors for their important work, and also for explaining how this type of method could potentially be used for medical time travel, which is a premise I totally agree with. Not surprisingly for a single paper, it alone has not resolved the key uncertainties about vitrification-based brain preservation.

The canonical answer is separate “side experiments” that will allow us to learn about how the nervous system works first, which will allow us to run those simulations. But what might those side experiments consist of?

A nice example comes from a new study using Patch-seq on close to 1500 human neurons expressing the neurotransmitter glutamate (i.e. glutamatergic neurons) across all six cortical layers. This method allowed the authors to measure the morphology, electrophysiology, and transcriptome from each individual neuron.

If we want to simulate a connectome, we need to know not just which neurons connect to which, but how those neurons function. Their use of Patch-seq allows us to build sort of a lookup table, so that given a neuron’s transcriptomic “type” (which we can infer from its morphology or biomolecular staining, and could be multidimensional), we can predict its electrophysiology.

Consider the action potential, which is the “spike” that neurons use to communicate (and the namesake of these monthly update emails!). There are a ton of aspects of the action potential that differ between neurons.

For example, when you inject current into different neuron types, some fire readily while others are more resistant. L4 IT (Layer 4 Intra-Telencephalic) neurons, which are the main recipients of thalamic input, need very little current to start firing. L2/3 IT and L5 ET neurons require around two to three times as much. This property, called the rheobase, was the single most distinguishing electrophysiological feature across neuron subclasses.

As another example, once a neuron starts firing, how does it respond as you push the current harder? That is captured by the F-I slope, which measures the rate at which firing frequency increases with more injected current. L4 IT and L5/6 NP (Near-Projecting) neurons have the highest F-I slopes, thus allowing them to relay graded input strengths. On the other hand, L6 IT neurons have nearly flat F-I slopes, meaning that they fire a burst at stimulus onset and then barely respond to further increases. This was the third most distinguishing electrophysiological feature between neurons.

This cell type-electrophysiology lookup table approach is very helpful. But part of the problem with it is that there may still be variation within cell types that also matters for their function.

The paper presents a nice example of this by zooming in on L4 IT neurons. Even within this single subclass, the action potential shape varies. In particular, there is a subtype called L4 IT_4 neurons, which have broader, slower spikes than the other L4 IT subtypes.

What explains this? They used a technique called nucleated patch recording, which involves pulling off part of the cell body, and which allows them to measure potassium currents directly. They found that L4 IT_4 cells have the least fast-inactivating potassium current (called A-type), while L4 IT_1 cells have the most. A-type potassium channels drive fast repolarization of the spike, so less of them means a slower downstroke and a broader action potential.

They found that the RNA levels of KCND2 and KCND3, which are the actual genes encoding the Kv4.2 and Kv4.3 channels that carry A-type current, predict the downstroke speed across the L4 IT population. But this expression varies continuously within the cell type, not discretely between types. So the transcriptome-derived cell type gets you pretty far, but there is still a spectrum of ion channel expression within each category that further shapes how individual neurons behave.

In this data set, it seems like the connectome plus morphologically-derived cell type classifications would probably allow you to predict a lot of the electrophysiological variance between cells. The remaining variance may require measuring, or somehow inferring, further molecular details such as the potassium channel expression. What degree of precision would be required for accurately simulating different aspects of cognition and behavior is still very much an open question.

2: A study uses artificial hibernation in mice as a new paradigm to study what structures are required to maintain long-term memories. Basically, they chemogenetically activate a subset of neurons in the hypothalamus to induce a hypothermic and hypometabolic torpor-like state for around 48 hours.

Because the subset of neurons activated are called Q-neurons, this is called Q-neuron-induced hypometabolism and hypothermia, or QIH for short. (It seems this is probably not possible in humans because we don’t have a clear analog of the Q-neuron population.)

During this artificial hibernation, the hippocampal firing rate drops by around 70% and the synapse density falls by more than 50%.

They next tested whether this loss of synapses led to a loss of long-term memories that were created prior to artificial hibernation. They found that these memories were not lost. The mice woke up with contextual fear memories and spatial navigation intact, and place cell representations maintained the same.

They then used two photon imaging to track individual dendritic spines through hibernation. They found that the dendritic spines that persisted maintained their original volumes.

And interestingly, the found that dendritic spines which were eliminated tended to regrow at the same dendritic locations within 24 hours. Specifically, they found 82.1% site-specific regrowth. This suggests that there might be some sort of structural scaffold at those dendritic locations that persists through the elimination and guides regrowth.

They next found that while overall engram synapse density drops during hibernation, spatially clustered engram synapses tended to be selectively protected. The non-clustered engram synapses tend to get pruned, while the clusters tend to survive. And these clusters tend to occur at multi-synaptic boutons (MSBs), which are axonal varicosities that synapse onto multiple postsynaptic partners.

There are a couple of possible interpretations of this study. The authors propose that a sparse “core memory trace,” which is defined by spatial clustering and multi-synaptic bouton connectivity, may be the critical aspect of long-term memory retention. In this case, non-clustered synapses would be dispensable. Alternatively, something about the hibernation state could enable network reconstruction upon arousal, with these clusters serving as seeds.

The finding that consolidated long-term memories can survive hypometabolism, hypothermia, and more than 50% synapse loss, as long as certain clustered synapses are selectively protected, adds to the existing evidence that long-term memory depends on key structural relationships rather than labile molecular states or ongoing electrophysiological activity.

3: Some of the wisdom of neuroscientist Adam Marblestone is captured in his recent podcast interview with Dwarkesh, which is mostly about how current AI models are similar and different to the brain. Here is my interview with Adam from 2015.

4: A new genetic method allows mapping of synapses to cell bodies using protein barcoding and light microscopy, which could be much more easily scaled than methods using electron microscopy. It is currently limited by a small number of epitopes (12) incorporated into the barcodes.

5: An argument that neuroscientists should study animals without neurons more often. This is because data has shown that multicellular organisms like Trichoplax adhaerens can still respond to neuropeptides. For example, they respond to endomorphin-like peptides in ways that suggest changes in their feeding behavior.

Neuropeptides operate via slower changes than electrophysiological ones, working primarily due to signaling via G-protein-coupled receptors. And yet, they alone are capable of mediating certain behavioral differences between animals.

To me this reminds me of the difference between a state and a trait. One might argue that the moment-to-moment variability in states between humans, such as whether one is hungry or thirsty, are not as essential for personhood as variability in traits such as long-term memories or personality. (More precisely I should say that at least that rings true for me, but I also think others should be free to choose what they value for themselves.)

6: A punchy argument by Nassir Ghaemi that mixed states (i.e. where manic and depressive symptoms are both present in the same patient) are crucial for understanding mood disorders. He argues that the DSM’s separation of bipolar disorder from major depressive disorder is wrong, that mixed states are highly under-recognized, and that this has clinical consequences.

He thinks this is especially the case in people with depression who also have baseline hyperthymic temperment, which manifests as constant high energy and a high degree of optimism. He notes that he thinks one of his friends had this condition, had it worsened with serotonin reuptake inhibitors (like Prozac or Lexapro), and then died of suicide.

As a result of this belief that mixed states are underrated, he now treats most of his patients with low-dose lithium (300-600 mg/d) rather than serotonin reuptake inhibitors.

Slightly too “me vs the world” for my taste, but it’s certainly a provocative essay and he might be onto something important.

7: An argument by Haim Belmaker, one of the pioneers of lithium treatment and research, that lithium should not be considered the gold standard treatment for bipolar disorder.

He notes that the superiority of lithium over alternatives has not been clearly demonstrated in meta-analyses. As alternatives to lithium, he sometimes uses valproate, carbamazepine, or combinations of the three. He says that he often has good results instead with low doses of second generation antipsychotics such as olanzapine 2.5 mg or risperidone 1-3 mg.

He also argues that despite decades of research, we do not really know the molecular mechanism of lithium. He notes that its proposed mechanism has conveniently shifted with academic fads in each era, from monoamines in the 1970s, second messengers in the 1980s, third messengers in the 1990s, and finally to neuroprotection in the 2000s. These findings were usually not replicated before being abandoned for the next fashionable framework.

I’m not sure I agree with him that the side effect burden of lithium is not much different than the alternatives. While this is a highly individual consideration, it seems to me to be meaningfully lower than some of the alternatives. However, because one of the major side effects of certain second generation antipsychotics like olanzapine is weight gain, this calculus has been changing rapidly in recent years with the new anti-obesity medicines.

8: An article about Mitul Desai, who after decades of helping manage his brother’s schizophrenia while building a successful career, quit to launch a company that works with insurance to provide coaching and peer support to caregivers dealing with mental illness in loved ones.

It’s a good summary of how caregiving for people with mental health problems is so different from other types of caregiving, due to there being unpredictable crises, isolation driven by stigma, and the impossibility of forcing treatment on adults.

9: A study of 85 participants ages 20-84 finds that chronological age is associated with lower reported baseline cognitive fatigue levels, while the “brain age difference” (how relatively old the brain appears relative to the chronological age based on MRI morphology) predicts how quickly cognitive fatigue accumulates during mental tasks.

A paper I was a part of previously found that histologically, brain age is most strongly linked to vascular structure. So to take a leap on my jumping-to-conclusions mat, I would hypothesize that the accumulation of cognition fatigue during tasks might be especially related to cerebrovascular function.

10: Merge labs, a new company planning to make brain computer interfaces that avoid implants and reach deep into the brain, has raised $252 million in funding. It seems they are primarily focused on using ultrasound devices.

11: A new study uses a portable device that does Raman spectroscopy on samples preserved in natural history museums to determine what type of fluid preservative was used:

12: A study finds that bats preserved in the fluid state for decades can still have the RNA content in their bodies analyzed precisely:

13: New-to-me data from 2025 shows that while religiosity in the US is down, belief in an afterlife is either fairly steady or actually seems to be slightly increasing. Here’s a plot of the data:

Belief in an afterlife is strong in all demographic groups, but is especially strong in women, Republicans, those who attend church frequently, and has an inverted u-shaped curve with household income:

It’s interesting to me that the dominant cultural narrative about death has not budged in the past 50 years, despite so many other changes in the world. If anything, it has gotten stronger.

14: An advance in scanning tunneling microscopy (STM), as a new paper introduces “inverted-mode STM.” Normally in STM, the probe tip manipulates or images atoms on a sample surface.

Here they flip it, so that specially designed molecules stand upright on a silicon surface, and as the probe scans over them, each molecule acts as a tiny imaging tip that maps the probe apex from below. This lets them verify the atomic structure of the probe tip while also enabling controlled chemical reactions at the junction.

They demonstrate the technique by reproducibly removing single hydrogen atoms from the probe’s silicon surface (27/28 success rate). The way this works is that a molecule with a reactive radical terminus, when positioned close enough, forms a stronger bond with the hydrogen than the silicon does, so the atom transfers to the molecule.

This is progress toward “mechanosynthesis”, i.e. using precise mechanical positioning, rather than applied voltage or heat, to drive chemical reactions and build structures atom-by-atom.

This is not my field, but as far as I can tell, this paper is a big deal. I’m surprised it has not gotten more interest. Robert Freitas and Ralph Merkle are co-authors.

15: Neuroscientist Peter Rupprecht’s annual “intuition about the brain” post includes a section on connectomics and its limitations for whole brain emulation. He distinguishes between current technical barriers (such as imaging scale and data analysis bottlenecks) and more fundamental questions about whether pure electron microscopy-based connectomics alone could ever suffice, even given future technological advances.

In the comments, he clarifies that he’s skeptical that connectomics without molecular annotation will ever be sufficient for reviving a person in silico, since neurons have variable distributions of ion channels and other proteins that electron microscopy alone cannot measure. But he notes that expansion microscopy with biomolecular labeling might eventually resolve these limitations. He concludes: “So, from the perspective of a brain preservation person, there is definitely room for hope for such future technology to reconstruct and revive brains in 100-200 years (although one can argue whether this preservation effort is justified).”

16: I was honored to see so many interesting comments on my post about the memory decoding award (see also on Reddit here and here, and Ken’s opening remarks on the award ceremony here). I wanted to signal boost and follow-up on part of Nicholas Kircher’s interesting comment here:

I have been playing around over the last year or so with the connectomics datasets from flywire, thinking of ways I might be able to simulate some structure or other, though my main limitation is that I don’t have data on what the expected result should be from a real specimen in terms of the neural firing patterns.

Ay, there’s the rub! That’s the exact problem. Nobody has great ways of figuring out what the expected neural firing patterns should be, and our ground truth data is thus far pretty sparse. For those who are interested in this area, here are a few avenues to consider:

A: Randal Koene has thought about this problem very hard, possibly harder than anyone else. He has created the Brain Emulation Challenge, which has synthetic electrophysiology data that modelers can try to predict.

B: So far the largest real-world data linking electrophysiology to connectomics I know of is in the MICrONS data set, which is from the mouse visual cortex. However, as far as I know, nobody has yet attempted to analyze this data to try to predict function from structure (as opposed to the other way around).

It would be interesting to develop benchmarks in this structure-to-function prediction, because we would expect our ability to do so to get better over time, as we learn more about the brain and can therefore infer more about the expected electrophysiology from the electron microscopy images.

C: People at Janelia, Harvard, and Google are working on a data set that will have combined calcium imaging and connectomics data from a larval zebrafish, called ZAPBench. This will probably be the best data set for benchmarking neural circuit models once it is released.

17: Aschwin de Wolf has started a substack, here is his first post.

18: Michael Cerullo has also started a substack. Here are his thoughts on Sebastian Seung’s recent podcast. In that podcast, when asked directly about the possibility of mind uploading, Seung deflected by framing it as a religious belief. Cerullo notes that this form of psychoanalysis is inaccurate, and it is more like a guilt by association argument. I agree.

19: BRLS is a non-profit that supplies $8-9 million yearly to cryonics research and has long been the major funding mechanism of the field. One of the best ways to figure out what the funded companies are up to is through Ben Best’s yearly updates, since he is the Director of Research Oversight at BRLS. Here is this year’s update.

One of the most exciting updates is the work at 21st Century Medicine (21CM) led by Ralph Spindler. He reports that they have done electron microscopy on rabbit brains, by opening the blood-brain barrier with detergent so that the cryoprotectant M22 can get into the brain. Work in this area from 21CM is reportedly being submitted to PLOS One.

It would be fantastic if another brain preservation protocol is able to achieve high quality structural preservation, especially if it can be done reliably in realistic conditions. More competition is good for the consumer.

I’m not sure what the methods were, and the available images aren’t high resolution enough for me to actually evaluate their data. I look forward to seeing the actual paper so that I can actually do so.

20: My talk at the December Sparks Brain Preservation open house is now online. The talk was about historical perspectives people have had on the topic of revival following long-term preservation, including those of John Hunter, Ben Franklin, Evan Cooper, Robert Ettinger, and Robert Prehoda.

The prize was launched in 2019. I previously discussed it in 2022. So far they have had three award ceremonies and panel discussions, each at a satellite event of the Society for Neuroscience conference, in 2019, 2023, and 2025.

A recording of the most recent panel discussion from last month is now public, and the transcript is here. I found it very interesting and have a few thoughts. But first, I want to give some background.

What is the current state-of-the-art (SOTA)?

Early anatomists around the end of the 1800s speculated that memories were stored in the physical structure of neural circuits. Ramón y Cajal, for example, called pyramidal neurons the “butterflies of the soul.”

In the 1960s and 1970s, environmental enrichment studies showed that different types of experience could lead to measurable changes in rodent neuroanatomy. This included increased cortical thickness and increased synaptic density. But that wasn’t specific to a particular memory.

In the late 1960s and 1970s, Kandel and colleagues established Aplysia californica as a more specific model for studying the cellular basis of learning and memory. They found that the sensitization and habituation of this animal’s gill withdrawal reflex led to functional changes in the strength of particular synapses in it.

In the 1980s, Bailey and Chen did the electron microscopy to show that these functional changes also entailed structural changes detectable after the ganglia were fixed. In particular, they imaged the presynaptic terminals of the sensory neurons that synapse onto motor neurons in the gill-withdrawal reflex circuit. They found that sensitized animals had more synaptic varicosities and active zones here, while habituated animals had less.

These studies by Bailey and Chen were landmark experiments for the field of neuroscience. They were, as far as I know, the first to demonstrate that specific memories correspond to structural changes at synapses. In principle, it seems to me that one could “decode” whether the animal had this reflex sensitized or habituated by examining these particular synaptic structures.

However, from the perspective of the Aspirational Neuroscience Prize, this involves only simple reflex memory, not the complex types of memories that are typically thought to define a person’s identity and experience. So, my understanding is that they would be considered a “trivial” form of memory.

While there have been studies since that have extended these findings to mammals and more complex learned behaviors, there has not yet — the Aspirational Neuroscience Prize argues — been a demonstration that the content of a complex, “non-trivial” memory can be read out from the structure of a preserved brain. As far as I can tell, the SOTA for extracting information from preserved tissue is still roughly at the level of Bailey and Chen from the 1980s.

What is a memory, anyway?

Back to the transcript from this year’s ANP panel. One of the main questions they discussed — as scientists are wont to do — is how the terms of the prize should be defined.

The moderator, Paul Middlebrooks, asked: “This is where I get confused about the difference between a behavior and a memory, because an alternate proposition was if you could accurately decode the last 10 bits of behavior of an organism before the sacrificing of the organism to fix its brain. How do you guys think about the difference between — Because a song is, in some sense, a memory, because you have to have learned it, but it’s also a behavior? Are those meaningful differences between a memory or a behavior?”

Randal Koene had a nice response: “I think when you’re talking about a behavior, the behavior, of course, is composed of a number of different steps, for instance, in that sequence of a song. Then you would say, ‘If I can differentiate between different pitches, different interval lengths,’ things like that, then you’re decoding something significant about the behavior that requires learning. Therefore, there is memory of some sort, right? Because memory, in essence, is anything where the future is dependent on the past. To me, that would be a memory as well, or it would be learned memory, if you decompose it into those parts of the behavior.”

Model-dependence

When we talk about information extraction, I am reminded of Michael Hendricks’s critique of cryonics from 10 years ago: “While it might be theoretically possible to preserve these features in dead tissue, that certainly is not happening now. The technology to do so, let alone the ability to read this information back out of such a specimen, does not yet exist even in principle.”

We need to be cautious here. When we talk about reading information or memory decoding, we can only really talk about interpreting imaged brain structure through the lens of a given theoretical model. This is similar to the Bailey and Chen results, where the researchers were able to decode whether the gill-withdrawal reflex was sensitized or habituated only because they already had a model of which synapses to examine and what changes to look for. Any memory decoding from a connectome will require substantial prior knowledge about how the particular brain circuit works.

As Randal Koene put it: “We have models of how these systems work. These are model organisms that have had decades of research on them. The very first thing that you’re doing is you’re saying here is our model of the hippocampus. Here is our model of how place cells and grid cells interact with each other. Now, given that model, which could be wrong, what is the experiment that we could do to try to decode the memory based on that model?”

What counts as “non-trivial”?

Andrew Payne asked about this directly: “I’m going to ask the question that I think needs to be asked, which is, non-trivial is doing a lot of work here. What does non-trivial mean?”

Nobody had a conclusive answer. It’s a difficult question to answer. One proposal came from Sven Dorkenwald: “This is why I think we should have this quantified in information-[theory] terms, so number of bits, and then we can argue about the number of bits that is non-trivial. It’s become a threshold problem.”

I agree with Sven it would be awesome to quantify this in bits. One reason is that it would let us track progress over time using a common denominator.

However, quantifying the information in an extracted memory in bits seems very tricky to me. Take the Bailey and Chen result. On the one hand, you could say they distinguished three states — habituated, control, or sensitized. But on the other hand, the underlying data could also be thought of as scalar. How many bits does that count as? It seems to me like it depends on (a) the precision of the measurement and (b) how reliably the structural differences are found to map to functional ones, both of which are difficult to be certain of.

Another potential issue with “non-trivial” is that it could be a moving goalpost. All the way until one can reproduce a particular ecphoric experience — which might require an emulation of nearly the entire brain — one could always argue that what has been decoded is not really a memory in the fullest sense. This is why it is useful to establish what people mean by this prospectively.

The zebra finch

A few of the panelists said that the encoding of songs in the zebra finch was perhaps the most tractable example of a non-trivial memory that could be decoded from a connectome in the near future. As Sven Dorkenwald noted: “I think the songbird is closest at this point, and I think it has the most well-established example of what would constitute a read out.”

Why is the zebra finch such an appealing model animal for addressing this question? A few reasons. First, the song they learn is undeniably complex. It is composed of different syllables each with specific frequencies and durations. Second, the learned song stays the same over time, which is helpful because you don’t have to worry about the memory changing between when you record the song and when you preserve the bird’s brain. As Michał Januszewski put it: “The song of the zebra finch is learned once for life. It doesn’t change. It’s crystallized.”

Third, there is a well-developed theoretical model for how the song is encoded. As Ken Hayworth explained: “There is a theory... that there is a chain of neurons in HVC that sequentially activate, like a timing board, and that those chain of neural activations are then feeding their axons down to a motor nucleus that gives particular tweets. The textbook model says that if that theory is correct, then this connectome will show what the bird learned.”

With part of the connectome of the zebra finch now mapped (discussed in last month’s update email), it seems we may soon be able to address this question.

Other potential model animals the panelists discussed were the mouse (possibly after learning a maze, but it’s unclear what the actual read out would be — maybe something in the hippocampus?), Drosophila (but there was some question if they even have non-trivial memories?), and the Etruscan shrew (the smallest mammal by mass).

When do the panelists think that the Aspirational Neuroscience Prize will be won?

Finally, the panelists also gave some predictions about when they thought that the challenge might be won:

Sven Dorkenwald: “I think this will happen once we can reconstruct the whole HVC region [in the zebra finch], which is, I think, possible now. I would say within the next five years, we should be able to see that.”

Michał Januszewski: “I would say the same, within the next couple of years, provided we actually start doing this and not talking about it.”

Randal Koene: “I’ll guess. I’m just agreeing. I think that, yes, the whole point is to just get started. I thought it was wonderful to just say, ‘Why not try?’ Yes, two to five years is my guess. I’m going to be really optimistic.”

Helene Schmidt: “I go with the five years.”

Andrew Payne: “What does non-trivial mean? … For example, we know in the [retinotectal] system that you can sensitize that to a moving bar, and you can read out the change in the compound synaptic currents from that. Probably, if you had the connectome, you could see that in that system. Have we already decoded that memory, or is that too trivial? It probably is, but what is the threshold? Can we get some discussion around that? Also, five years.”

Paul Middlebrooks (moderator): “I feel like I could make a lot of money from you guys if we just place a wager about this, because my guess is I’m much more pessimistic than all of you. It’s awesome, and why not try and guess, rainbows and all that great stuff, but I would guess 30-plus years.”

I created a manifold market on when the Aspirational Neuroscience Prize will be won, if ever before 2045. My personal guess would be somewhere around 2035, but it also depends heavily on how rapidly AI capabilities progress. That’s because many of the bottlenecks, like proofreading ultrastructural data (currently thought of as the biggest bottleneck), annotating the structures with physiological knowledge from the literature and side experiments, and iteratively testing candidate decoding models, could all plausibly be accelerated with better AI systems.

How is this relevant to brain preservation?

Memory decoding is cool and all, but to be honest I probably wouldn’t care that much if I didn’t think it was related to brain preservation.

As you recall, the Aspirational Neuroscience Prize was launched in 2019 by the Brain Preservation Foundation.

One of the reasons I was super interested in it is that the Brain Preservation Foundation prize — a challenge to demonstrably preserve synaptic connectivity across a mammalian brain — came and went (launched in 2011, won in 2016/2018) without a super major uptick in interest in brain preservation.

It seemed like part of the reason why people weren’t all that interested in the Brain Preservation Foundation prize might have been that they didn’t agree that if the synaptic connectome was preserved, it would mean that the information for long-term memories would very likely be preserved as well.

Ralph Merkle had already described the Bailey and Chen results in his article “The Molecular Repair of the Brain” as evidence that memory is encoded at the structural level, and as part of his justification for structural preservation in cryonics being sufficient. But maybe people wouldn’t consider the Bailey and Chen result sufficient because it is “just a reflex” and that “real” memories might not work this way.

(More likely, most people just don’t know about this literature. But I do think it’s reasonable for some people to view them as not directly relevant, because they involve only reflex learning. Even though my guess is that complex memories are also “just reflexes” structured in a more layered and hierarchical way, the proof for this is circumstantial.)

To quote from the Aspirational Neuroscience website: “The current neuroscience consensus is that the vast majority of long-term learning and memory is encoded in the patterns and strengths of synaptic connections among neurons. But there remains some doubt because most experimental evidence is indirect. This challenge, to decode a memory from connectivity alone, represents a powerful test of the quality of our neuroscience theories of learning and memory.”

Two to five years sounds optimistic to me, but I’m obviously rooting hard for them to succeed. If this prize is won, which I do expect will eventually be possible with tissue preserved using today’s methods, then I think it would get more technical people interested in brain preservation. The discussion about what type of brain structure needs to be preserved to retain the information content of long-term memories could become more concrete and less theoretical.

A new paper proposes another alternative: photoemission electron microscopy. This method is commonly used in materials science. It uses UV or X-ray light to eject electrons from a sample surface, which are then imaged to reveal nanoscale structure.

Using laser illumination, they could resolve the plasma membranes of individual neurons even at exposure times as short as 100 ms (A), 10 ms (B), or 1 ms (C):

They note that because they do the illumination with photons instead of electrons, it is possible that different sample preparation methods could be used to increase contrast.

They estimate that a single PEEM at current speeds would take 8.2 years to image the connectome of an adult mouse brain. But they note that faster detectors and parallelization across multiple instruments could reduce this dramatically.

2: Of the around 500 million neurons in an octopus, 300–350 million are in their arms. Their arms function somewhat autonomously — i.e., they can sense the environment and control movements without waiting for instructions from the brain.

A new study does calcium imaging in ex vivo slices of octopus arms. They find that glutamate and dopamine are the major excitatory drivers in arm ganglia. Notably, serotonin was found to lower the subsequent excitability of the circuit, dampening the response of the circuit to excitatory signals, but only when it was applied before the excitatory transmitters.

3: What’s in a brain region? A new study applies a parcellation method based only on geometry to the brains of different mammalian species. They argue that their method better captures the developmental process of brain region formation than existing approaches. I’m not sure if that’s true (it’s not my area), but I found this to be a beautiful figure:

4: A conversation with neuroscientist Tatiana Engel about connecting low-dimensional neural manifolds to the underlying circuit connectivity through “latent circuits.” One thing they discuss is the International Brain Lab, which generates neuropixel data, and has a goal to share the data with the public, even prior to publication.

5: Thomas Reilly on “The return of psychiatric eugenics”.

6: Sublingual dexmedetomidine as another medication option for patients with agitation.

7: It turns out that Oliver Sacks was lying about a lot of his cases. Especially in “The Man Who Mistook His Wife for a Hat,” his most interesting book and the one that made him famous. It is a bad situation for humanity for many reasons (although it is obviously good that it was discovered). An obvious parallel is Jonah Lehrer, but this is way, way worse. (I personally thought Lehrer’s failings were not all that bad and felt badly for him.)

8: Steve Byrnes on whether AI systems will have what he calls “approval reward” — basically, the innate drive of humans to seek the approval of others, which makes us intrinsically motivated to follow norms rather than act as ruthless consequentialists. Seems cruxy.

9: In a win for body autonomy, Medical Aid in Dying passes in New York and Illinois.

10: “Reframing Formalin”: An article on how formalin fixed biospecimens in historical collections can allow for retrospective gene expression studies. They note: “Thus, exposure to formalin essentially corresponds to taking a snapshot of organism‐wide gene expression at the time of death.”

11: This article on a swine model of isolated eye perfusion for transplant has a nice figure of a perfusion circuit setup, showing the plastic reservoir where solution is stored (1), an oxygenator (3), bubble trap (5), pressure monitor (6), and a machine where Blood Gas outflow samples are analyzed (8):

12: Max Hodak’s Science Corporation is starting a new division of the company focused on organ preservation via ex vivo perfusion, starting with rabbit kidneys.

13: A new company, Netholabs. They are training neural foundation models on massive paired neural-behavioral data from freely-moving mice with wireless brain recordings. Their goal is “synthetic organisms for neural disorder and whole-brain-emulation science.” Here is a talk by co-founder and friend of the blog Christian Larsen.

14: Another new company, Memazing, this one led by neuroscientist Sebastian Seung. They are taking detailed connectome maps (starting with the fly brain’s 140,000 neurons and 55 million synapses) and simulating them in software. They are planning to use synaptic connection counts as estimates for connection strengths and then apply ML techniques like backpropagation to tune unknown parameters (like excitatory/inhibitory balance) until the simulations produce behavior that matches real experiments. “Memazing” is an interesting name for the company. Here is a podcast Seung did with Ashlee Vance about it:

Here is Ken Hayworth’s take, which I totally agree with:

15: Possibly related, the Institute for Advanced Study is offering postdoctoral and sabbatical memberships for a 2026-2027 special year program on Modeling Fly Vision, co-organized by Misha Tsodyks and Sebastian Seung, that will bring researchers together to model the Drosophila visual system using connectomics and transcriptomics data. Like many others (I would bet), I have long thought it is promising to use transcriptomics to identify cellular parameters for circuit simulations based on a connectome.

16: New blog by cryonics pioneer Mike Darwin. Here is one of his first posts, “What Fresh Hell is This? The Rise and Growth of Antinatalism and Promortalism”.

17: Interview with Hunter Davis, CSO of Until Labs. Describes their plans for donor organ vitrification. He points out the difficulty of preserving full neural function in a brain. He also notes: “And I think initially we really were committed, and still remain very committed to this day, to this hibernation as the long-term goal. And this is the idea of taking an entire person, taking an entire patient who would be terminal with some disease and giving them access to cures that are right around the corner.” However, he suggests that achieving long-term human suspended animation in this manner is not a near-term goal of theirs at all.

18: This write-up by Nikki Olson and Aschwin de Wolf describes a tour of the Sparks Brain Preservation facility this month. Thank you so much to them for organizing the event and their generous coverage.

Biostasis Pacific Northwest

Sparks Brain Preservation: Clinical Excellence and Operational Readiness

On Saturday, December 13th, at noon we visited Sparks Brain Preservation for a facility tour. The site is located about 50 minutes south of Portland in Salem, Oregon, and is easy to access. The facility sits on a large property alongside multiple other buildings owned by…

Read more

7 months ago · 5 likes · Nikki Olson and Aschwin de Wolf

19: A preprint of an article I co-authored with Ariel Zeleznikow-Johnston and Emil Kendziorra is online: “Physician estimates of the feasibility of preserving the dying for future revival.” In it, we surveyed 334 US physicians about their opinions on the feasibility and ethics regarding human preservation procedures.

Ariel has a very nice summary here. My personal favorite part of the paper is the text comments from the physicians (which were optional), available for download as an excel file here. Perhaps one day I will go through some highlights of that.

One of the interesting aspects of this paper is that it is the first time an article I have contributed to has been discussed outside of academic circles. The results… well, maybe I’m okay with sticking to academics.

For example: Crémieux is a prominent Twitter user with 283k followers. Here was their take:

I find it kind of hard to describe how many things are wrong about this tweet.

A: It’s obviously not accurate that a 25.5% median probability estimate means doctors “don’t believe” something works. That’s like saying “Doctors do not believe in CPR for cardiac arrest. It only has a 22% survival-to-admission rate, therefore it doesn’t work. When your heart stops, you’re dead.” A one-in-four estimated probability for a speculative procedure with no heretofore empirical validation is actually pretty high IMO.

B: The question specified an ideal preservation outcome, where “imaging and brain biopsies show intact brain structure down to the synaptic level, including the spatial distribution of key biomolecules.” This is an ambitious target, which has never been demonstrated across a single preserved human brain, let alone being shown to be reliably achievable. Being “frozen” (i.e., undergoing ice crystal damage) would be incompatible with this preservation scenario.

C: Obviously I think medical doctors are an interesting group to survey, but they clearly are not the final authority on neuroscience or information theory. Even if they had said 0%, it wouldn’t mean the information is actually destroyed. It would mean doctors are uniformly very skeptical. The logical leap from “physicians have a median estimate of 25% probability” to “you die and the information encoded in your brain is lost” is absurd.

In sum, this is just a horribly misguided take. It would be great if people read papers or abstracts before tweeting about them, or at least had their LLM of choice do so.

A new study generates the first connectome of Area X, which is a specialized nucleus within the larger basal ganglia system of zebra finches that is responsible for learning the courtship song. Because Area X contains all the major basal ganglia cell types (i.e. striatal, pallidal, and subthalamic neurons) in one small region, it also offers a tractable way to study an entire basal ganglia circuit at synaptic resolution.

To preserve the zebra finch brain, they fixed it via perfusion with formaldehyde and glutaraldehyde at high pressure. They then did serial block-face electron microscopy to generate a volume of 256 x 256 x 384 µm3. They primarily used automated reconstruction and found that it was quite effective even without extensive proofreading, allowing them to eventually have 8,576 fully reconstructed neurons:

In vertebrates, the three basal ganglia subcircuits usually described are the direct, indirect and hyperdirect pathways. They were able to trace these pathways and also to draw quantitative conclusions about their relative strengths — i.e., the percentage of synaptic area per pre-synaptic/post-synaptic cell type. Here is that data for the hyperdirect pathway:

This is super interesting for people studying the basal ganglia, but maybe not for some other people. In a companion paper, a group uses this connectome to study a question of more general interest: the biological solution to the credit assignment problem in machine learning.

But first, what is the credit assignment problem?

As we take actions and experience things, it is useful if we can change our nervous systems to become better adapted to our environments. This is true of both artificial and biological neural networks.

In artificial neural networks, researchers usually use backpropagation for this. This is a famous algorithm that propagates error signals backward through the layers of a neural network, calculating gradients to determine how each connection weight should be adjusted to minimize prediction errors.

Backpropagation has been critiqued as not biologically plausible. Instead, an alternative algorithm called node perturbation has been proposed. In this model, instead of computing exact gradients, the network estimates gradients by injecting random noise into neurons and measuring how these perturbations affect performance. This allows each synapse to learn based on the correlation between its own activity, the injected noise, and a global reward signal:

How could this be implemented in the brain? To understand, we need some more anatomical background. The Medium Spiny Neurons (MSNs) in the zebra finch Area X receive synaptic input from three key brain regions, which are each thought to have different roles:

a: Lateral magnocellular nucleus (LMAN) - this acts as a source of vocal variability in the song motor system.

b: HVC (a brain region whose name doesn’t stand for anything) - this generates a sparse timing signal that controls vocal output at each moment in the song.

c: Ventral tegmental area (VTA) - this sends information about song performance via dopaminergic axons.

The authors argue that the node perturbation model makes an anatomical prediction: if neurons from the LMAN inject variability (like adding noise) and those from the HVC encode timing context, then LMAN axons should preferentially synapse onto MSN dendritic shafts, while HVC axons should preferentially synapse onto MSN dendritic spines.

This is because spines are electrically isolated compartments well-suited for coincidence detection and maintaining temporary eligibility traces, whereas shaft synapses can depolarize the dendrite more broadly to modulate plasticity at a set of nearby spines.

And indeed, this is what they found in the Area X connectome data set. HVC axons synapsed predominantly onto dendritic spines, whereas LMAN axons showed a preference for dendritic shafts:

In other words, this is evidence that credit assignment could be solved in biological neural networks based on having different types of excitatory synapses interacting on a single dendrite: one type to inject variability (in this case, LMAN axons), one type to correlate that variability with reward signals (timing context from the HVC), and one type to assign credit to the synapses that lead to improved performance (dopamine reward signals from VTA).

They note that the way credit assignment is implemented might be different in different brain regions. But just due to parsimony, I kind of doubt that. My guess is that whatever the learning algorithm is, it is probably the same in different brain regions, at least for spiking neurons. However, while their results are compelling, they’re still largely correlational, and this paper is not the final word on the topic.

2: A study in mice uses correlated light and electron microscopy to map the synaptic inputs onto the dendrites of three interneuron types — those containing calbindin-D28K (CB), calretinin (CR), and parvalbumin (PV). 3D reconstructions show that parvalbumin-positive interneurons have the highest synaptic density and also are the only type with clear spatial clustering of synaptic inputs.

3: Evidence in the blood of elevated levels of the protein Myelin Oligodendrocyte Glycoprotein (MOG) around 7 years before a diagnosis of multiple sclerosis (MS). 1 year after that (around 6 years prior to diagnosis), they found elevated levels of neurofilament light chain, indicative of axon damage.

4: Study uses thousands of histological sections from 5 formalin fixed human “hemibrains” (one half of a brain) to develop a new probabilistic atlas of the brain that can be used for MRI studies:

5: A study finds that intact whole rat brains stored in formaldehyde at 4°C have stable c-Fos and GAD67 immunofluorescence signals for up to 10 months with no degradation. However, pre-cut brain slices stored under these same conditions showed a loss of c-Fos signal and a slightly increased GAD67 signal. My guess is that the loss of c-Fos staining is likely due to ongoing crosslinking causing epitope masking of this protein, which is present in the nucleus (a crowded space).

6: A study demonstrates a new method of “atomic stencilling.” They first create iodide masks and then “paint” polymer coatings over these to create patchy particles in precise patterns. This could be thought of as another baby step in the very long road towards generalized nanotechnology.

7: A new method of drug delivery: gel capsules with iron oxide nanoparticles that can be steered through blood vessels using external magnetic fields. The nanoparticles can then be heated by a high-frequency magnetic field to dissolve the gel and release the medication at the targeted site.

8: A new trial (n = 62) finds that IV ketamine is not more effective than IV midazolam for the treatment of depression in an inpatient setting. I’ll note that benzodiazepines are also fairly effective for depression short-term though.

9: Delix Therapeutics was founded in 2019 to develop psychoplastogens (aka neuroplastogens), which is a term for compounds designed to rapidly promote neuroplasticity. They just reported Phase Ib trial results showing that zalsupindole, a non-hallucinogenic analog of 5-MeO-DMT, led to a ~50% reduction in depression scores in patients with depression, which persisted for 4 weeks after treatment ended. The FDA has now cleared a Phase II trial design with at-home self-administration of zalsupindole.

10: Alzforum discussion of three new phase I clinical trials of stem cell replacements in Parkinson’s disease. BlueRock is conducting a phase III trial to test the efficacy of stem cells to replace dopaminergic neurons, with a plan to enroll 102 participants with moderate severity Parkinson’s disease.

11: Genetic correlations between classically neurologic and psychiatric conditions. Some interesting findings, such as a negative correlation between OCD and stroke risk. Others I expected from previous data, such as there being a positive correlation between general cognitive ability (COG) and autism spectrum disorder (ASD), anorexia, and (insignificantly) OCD, as opposed to a negative correlation between general cognitive ability and most other psychiatric diagnoses. I love looking at this kind of correlation matrix. I encourage everyone to just take their data and throw all of it into a big correlation matrix.

12: There are now more personalized gene-editing therapies for rare diseases, such as CRISPR. A new paper discusses efforts with the FDA to establish a regulatory pathway for “umbrella” treatments that could treat patients with different genetic variants under a single platform approval.

13: A history of how the electron microscope was invented.

14: Elon Musk predicts mind uploading in 10-20 years. He argues this could help address AI alignment by increasing human cognitive bandwidth, the concern being that if humans communicate too slowly relative to AIs, AI systems might simply ignore us. And Elon has famously never been too optimistic on technology timeline predictions.

15: A very nice report on the current state of whole brain emulation by Niccolò Zanichelli, Maximilian Schons, Isaak Freeman, Philip Shiu, and Anton Arkhipov. Worth checking out if you’re interested in this space.

They have a chart of the decreasing cost of mapping a single neuron over time:

They also have a nice figure showing the currently available electron microscopy-based connectomes for different organisms:

And here they have a diagram of some of the components it would take to build an emulation of C. elegans:

16: A presentation by Jordan Matelsky at the Foresight Neurotech Workshop on the path to scaling whole brain emulation. Argues that structural imaging is scaling faster than functional recording, so the key bottleneck is now solving the “structure-to-function” problem. I fully agree with this. And that perturbation experiments can dramatically reduce the data requirements for learning how the molecule-annotated connectome maps to neural transfer functions. I’m unsure on that part, but it seems reasonable.

17: Episteme is a new for-profit R&D company that aims to do research in physics, biology, and engineering. If you look at their jobs board, two of the jobs are for neural circuit imaging and modeling:

The job description says that they will be working with Michael Skuhersky, who recently wrote an article on the importance of simulating C. elegans brains.

This suggests to me that one of their areas of focus may be on simulating the nervous system of C. elegans.

In other words, it looks like Eon Systems might have some competition in the for-profit world of brain simulation and emulation R&D.

It’s hard for me to see what near-term applications this type of R&D would have, unless one expects very rapid AI development that would accelerate progress dramatically. However, there are quantum computing companies even though some people estimate that useful quantum computing (compared to classical computing) is still many years away. So maybe it makes sense that there would be neuro simulation start ups as well, making a similar bet.

18: A couple of weeks ago I had a moderated discussion with Aschwin de Wolf about comparing fixation and pure cryopreservation as two different approaches to brain preservation. Thanks so much to Becca and Max for organizing this and including me in it.

Expansion microscopy — chemically expanding the tissue and then imaging it with light microscopy — has emerged since the mid-2010s as an option for improved resolution without the high cost of electron microscopy.

A fascinating new paper from the team at E11 Bio introduces a method to combine expansion microscopy with a clever molecular barcoding system to map the brain. They inject viral vectors (AAVs) carrying 18 different protein “bits” — i.e., variants of GFP tagged with distinct epitopes — that cells then take up and randomly express in different combinations:

As a result, each neuron that takes up the virus gets a unique molecular barcode (theoretically up to 262,143 combinations), which is expressed throughout the cell.

To preserve the mouse brains, they perfused them with formaldehyde, removed them from the skull, and post-fixed them under refrigeration for 28 hours. After using the Magnify expansion method and other forms of processing, the barcodes can then be read out for iterative rounds of immunostaining and imaging:

One limitation with their current barcoding approach is that it only labels a subset of neurons. In the hippocampal dataset, only 49% of cell bodies (146 out of 298 somas) expressed at least one protein bit. Among those labeled cells, 80% had unique barcodes, so one in five labeled neurons also shared its barcode with at least one other cell.

This sparsity is due to their stochastic AAV infection approach. Each neuron independently takes up some random subset of the 18 different viruses in the pool, leading to combinatorial diversity but incomplete coverage.

This makes their approach different from electron microscopy or dense light microscopy methods, which aim to reconstruct every neuron in a volume. As it currently stands, their method is better suited for applications requiring partial reconstruction, rather than building complete wiring diagrams.

Using this approach on 5-fold expanded mouse brain tissue, they were able to reconstruct individual neurons across a 10 million μm³ volume of the hippocampus, with each neuron assigned a different pseudo-color based on its unique barcode. This led to a map around 9 million synapses throughout the tissue, which were identified by the co-localization of presynaptic (Bassoon, cyan) and postsynaptic (PSD95, magenta) markers:

Here is an example of two reconstructed neurons with their morphologies traced. The inset displays a higher-resolution view of neuronal processes, with the axons and dendrites labeled to identify putative synaptic connections. Note that axons tend to be thinner than dendrites, which is also seen here:

Because they use light microscopy, they are also able to relatively easily stain for particular biomolecules, which is a huge advantage over electron microscopy, where this is possible but more difficult. Here they show staining for different biomolecules at synapses, which can be seen in highly localized sub-cellular positions:

The senior author Andrew Payne notes in an interview that the mission of the company is to eventually enable the mapping of the whole human brain.

Because they use viral vectors to express the barcodes, the options for human application seem relatively limited. Injecting viral vectors into living human brains purely for research purposes seems pretty much out of the question (?). And keeping large volumes of postmortem tissue viable long enough for the viral vectors to express and the barcodes to fill the cells seems difficult.

Another option would be to try to identify endogenous molecular barcodes, i.e. unique combinations of naturally expressed proteins, RNA transcripts, or other molecular markers that already distinguish individual neurons from each other to a sufficient degree of accuracy. If enough natural variation exists between cells, they might be able to use these innate molecular signatures for tracing without needing to deliver artificial barcodes.

One particularly promising candidate are protocadherin proteins, which appear to create locally unique expression patterns in each neuron, allowing neurons to distinguish self from non-self when making connections. The main problem is probably copy number, as protocadherins do not seem to be nearly as abundant as the proteins virally expressed in this paper. So other candidates or combinations would probably need to be identified.

I am especially interested in the endogenous molecular barcode option because it is one method I believe could be used to infer the original morphological states in the presence of damage following brain preservation. So I would really love to see this idea of mapping endogenous cellular barcodes explored more.

2: The complete connectome of an male entire Drosophila central nervous system has now been published. It contains 166,691 neurons.

Not sure how this fits into previous Drosophila connectome efforts? Don’t worry, the authors compare it to a few of those — one from a female brain (“Flywire”), one from a hemibrain, and one from a Male Adult Nerve Cord (MANC). All of these connectomes were highly similar. They were able to match 96.4% of central brain neurons to neurons in at least one of the previous connectomes.

Interestingly, they were able to identify aspects of sensory neurons — both which nerve they enter through (antenna, leg, wing, etc.) and their modality (vision, taste, smell, touch, temperature, humidity, etc.) — based on anatomical knowledge of nerve organization, without needing to preserve the peripheral receptors or body parts themselves. Similarly, motor neurons could be catalogued based on their exit nerves and muscle targets inferred from anatomical knowledge. This is an example of how the essential information for brain functioning, from sensory processing to motor control, is contained within the CNS alone:

3: A new comparative connectomics study compares two Drosophila mushroom body data sets prepared and imaged separately by FIB-SEM. They found that the convergent synaptic motif, which is the name for a microcircuit in insects in which two nearby axons form presynaptic sites onto the same postsynaptic neuron, was present repeatedly across individuals. They suggest that that it may be a conserved building block of neural computation:

4: What happens to human synapses immediately after electrical stimulation? A new study uses acute brain slices from epilepsy surgery patients, i.e. freshly resected tissue that is kept viable after removal for a short period of time, to study this for the first time in humans. They find it is basically the same as in mice.

Before electric field stimulation, there are normal resting synapses with synaptic vesicles. 100 ms after stimulation, you can see uncoated endocytic pits (i.e. membrane invaginations that enable synaptic vesicle recycling, indicated by black arrowheads). By 1 second, you see mostly endosomes (i.e. membrane compartments, indicated by black arrows), which can remain for at least 10 seconds:

5: A new study uses whole-brain astrocyte Fos tagging in mice during a fear conditioning task. Fos is a protein commonly used as a measure of neural activity. They find that astrocytes play a role in providing memory stabilization after recall.

The process involves two steps. First, the initial memory formation upregulates adrenergic receptors over around 24 hours. Second, subsequent recall triggers these primed astrocytes to express Fos and secrete the protein IGFBP2, which they show stabilizes the memory. The authors were able to visualize these multi-day molecular traces in astrocytes following aldehyde fixation:

6: A new study characterizes a poorly understood form of direct communication between neurons called dendritic nanotubes. These are long, thin, actin-rich bridges that connect dendrites to one another directly rather than through synapses. They were able to visualize them after fixation and brain clearing:

They found that these nanotubes allow for the spread of ions and molecules between connected neurons. This enables them to propagate calcium signals and also to facilitate the spread of pathologic molecules such as amyloid-beta:

7: A new study uses genomic data from ~500,000 sibling pairs to estimate heritability for 14 traits. The key innovation is the use of a within-family design, which compares siblings who share the same family environment but randomly differ in how much DNA they inherited from each parent.

Traditional between-family studies compare unrelated individuals. The within-family design allows them to mitigate shared environment confounding (e.g. parenting, diet, socioeconomic status), avoid population stratification (i.e. ancestry-related environmental differences), and separate direct genetic effects from “genetic nurture” (i.e. the environments that parents create based on their own genes). They found some differences using this method. For example, they found that the heritability of type 2 diabetes was 0.43 (within-family) vs 0.62 (between-family), suggesting that previous heritability estimates were inflated by shared family lifestyle.

It’s an elegant design but requires a large sample size. Enter 23andMe. (It’s a general rule that most genomics data is not really that useful to a person on its own, but it only becomes useful when aggregated in very large data sets.)

Of the traits they studied, I found “Talk vs listen” to be the most interesting, as they found that it is very heritable (0.48) and I haven’t heard much about it or related traits before. It was based on a question that 23andMe users were asked: “If forced to choose, would you say you generally prefer to talk or to listen?” (0: Talk, 1: Listen, NA: I’m not sure). It was found to be more heritable than even type 2 diabetes or asthma.

I guess the takeaway is that the next time someone says that you are talking too much you can say it’s not your fault, it’s just your DNA.

8: New genomics study from more than half a million Taiwanese residents of primarily Han Chinese ancestry with matched electronic medical record data.

I was particularly interested in the correlations among genetic risk scores they found for different phenotypes. In particular, they found that there were basically three genetically correlated clusters of conditions: (a) “metabolic syndrome” (e.g. obesity, high cholesterol, diabetes, high blood pressure), (b) “autoimmune diseases” (e.g. Sjögren’s syndrome, lupus, psoriasis, thyroid disease), and (c) kidney-associated conditions (e.g. chronic kidney disease, kidney stones, gout, ankylosing spondylitis):

9: Abhishaike Mahajan on the moral case for working in biology. My comment pointed out that while I fully agree with this, such exhortations must be balanced against the practical difficulties of sustaining a career in biomedical research. Plenty of people would like to work in biomedical research but find that they effectively cannot do so for reasons outside of their control.

10: Related: Career advice from Thomas Insel, a former director of the NIMH. He points out that a lot of scientific research these days (2025) is done at for-profit companies.

11: Randomized trial of the investigational new drug seltorexant for insomnia (n = 364, 2 week trial).

One of the key outcome metrics was WASO-6 — basically, the total minutes spent awake during during the first 6 hours after the person initially fell asleep. You want this to be lower (obviously). They found that seltorexant had better WASO-6 outcomes compared to placebo (around 30-40% improvement) and compared to zolpidem (Ambien) at 2 weeks (around 30% better), whereas zolpidem’s initial efficacy at day 1 waned.

Seltorexant is a selective orexin-2 receptor antagonist. Compared to dual orexin receptor antagonists (DORAs) that also antagonize orexin-1 receptors, they argue that the selective orexin-2 antagonism will help preserve non-REM sleep. It would have been nice to see a direct comparison with a DORA (a few of which have been approved), but it’s understandable they wouldn’t want to risk not seeing any improvements.

12: A capacity evaluation is a clinical assessment of a patient’s mental ability to make their own medical decisions. At least in the US, this is commonly thought of as an aspect of psychiatric expertise. There are existing frameworks on how to do capacity evaluations, but not as much guidance on whether to do them.

Friend of the blog Jacob Appel has written a new article with a three step guide for making that decision. First, to assume capacity exists without evaluation unless there is reasonable uncertainty. Second, to determine whether the results of the evaluation would impact patient care anyway. And finally, to weigh the potential benefits of the assessment against its costs in terms of the patient’s well-being.

13: Jan Betley reports having isolated anhedonia while taking semaglutide that improved upon decreasing the dose.

14: An argument that valproate (Depakote) should have REMS regulation for women of reproductive age due to its very high risk of teratogenicity.

15: Nolan Williams was famous in the psychiatry community for many reasons, including studies of accelerated TMS for depression (such as the SAINT trial). He recently died. Owen Scott Muir wrote a eulogy for him.

16: Tomás Bjartur wrote a post on the technical feasibility of molecular nanotechnology. Points out the difficulty of proving that it would not be possible:

“The design space is very large. Any proof that Drexlerian nanotechnology is impossible would have to be very robust. And it strikes me that most plausible looking arguments can likely be hacked around given how large the design space is.”

A good comment from SE Gyges: “[I]n every case people tend to think that things are possible only when they know more exactly how they will be done.”

17: Start-up Bexorg raises 23M for its technology using postmortem brain perfusion and profiling as a method for drug discovery.

18: Are you going to be in San Diego in mid November, for example going to the Society for Neuroscience (SfN) conference? If so, you might consider going to the Aspirational Neuroscience Awards event on November 17 at 7 pm PT at the Hilton San Diego Bayfront. Here are the previous award winners. Here is the previous panel discussion from 2023. To RSVP and learn more about this year’s event, see here.

19: On Nov 13th at 11 AM PT, I will be having an informal discussion with Aschwin de Wolf on the Cryosphere Discord on the differences between fixation and pure cryopreservation. Here is the invite link.

20: Charles Platt, author and former Director of Suspension Services at Alcor, came to visit us at Sparks Brain Preservation and wrote about our organization. Discussion on Reddit here and here. Here is his Substack post, which is quite interesting (at least to me) and has a few plot twists:

The Biostasis Standard

An Alternative to Cryonics

In 2002, a dentist in Oregon named Jordan Sparks felt so dissatisfied with cryonics, he decided to start his own organization…

Read more

10 months ago · 12 likes · 9 comments · Charles Platt

It was wonderful to meet Charles in person. We went to lunch in Salem. I ordered an Impossible burger and french fries. He pointed out that my meal was “not very life extensionist.” I was forced to concede the point. I could only nod gravely and continue chewing.

21: Posting on NN may be light for awhile as my wife and I just welcomed our second child. Thank you as always for your support of this newsletter.

First, “mossy fibers” carry information from diverse sources (the pons, spinal cord, etc) into the granular layer. There, tiny granule cells, which are the most numerous type of neuron in the brain (there are roughly 50-70 billion in humans), help to transform these inputs.

Each granule cell has a T-shaped axon that rises vertically and then splits to run for millimeters as a “parallel fiber” through the molecular layer. These parallel fibers intersect with the elaborate, single-plane dendritic trees of Purkinje cells. Finally, the Purkinje cell axons are the sole output of the cerebellar cortex, sending inhibitory projections to the deep cerebellar nuclei.

A new connectomics study captures how this wiring emerges during mouse development. They reconstructed cerebellar cortex circuits from brains preserved at five timepoints during the first two weeks of postnatal life (postnatal day 0 (or P0), P3, P7, P10, and P14).

The number of parallel fibers with connections to a Purkinje cell increased over time, from 64 connections at birth, to 271 at day 3, 742 at day 7, 18,195 at day 10 (an especially dramatic increase), and 34,178 at day 14:

The increase in parallel fiber inputs to Purkinje cells coincides with the progressive organization of Purkinje cells (PCs). By day 10 (panel D) and day 14 (panel E), the Purkinje cells have achieved their mature configuration, i.e. being aligned in parallel planes with highly organized dendritic trees. This figure shows adjacent Purkinje cells in different colors to demonstrate this:

Part of the mechanism behind this massive synaptogenesis can be seen on day 7, where Purkinje cells are found to have thousands of transient dendritic filopodia. These filopodia are long, thin projections extending perpendicular from the dendrites and parallel to the direction of parallel fibers. Compared to dendritic spines, they are longer and lack a head structure at the end.

They found that these filopodia are sites of active synapse formation. The colored patches in the figure below (purple, green, red) indicate synapses with different maturity levels based on their synaptic vesicle counts.

By day 10 and day 14, many of the dendritic filopodia have been replaced by dendritic spines. The 18,700 spines at day 10 and 35,748 at day 14 are shown in random colors. These spines maintain the same parallel fiber orientation, creating an organized array of synapses when combined with the parallel fibers (a subset of which were traced and are shown in yellow).

Despite the dense packing of parallel fibers and adequate opportunity for redundant connections, the neighboring Purkinje cells were found to share significantly fewer parallel fibers than expected by chance. Instead, each Purkinje cell develops a largely individualized cohort of inputs through selective synapse formation. This likely helps them to process different types of information.

Overall, it’s not really surprising that aldehyde preservation methods are able to preserve all of these structures so precisely, nor that we can reconstruct complete circuits from preserved tissue using enough volume electron microscopy. Instead, I think of this study as a nice example of how connectomics at different developmental time points can reveal principles of how the nervous system is initially wired.

2: Another developmental connectomics study, this time on the Drosophila larval nociceptive system:

They have a nice diagram showing how they can convert annotate raw 2D electron microscopy data and turn it into 3D tracings, in this case for a post-synaptic density:

The found that from the early larval stage (L1, around day 1) to the late larval stage (L3, around day 5-6), dendrites in this circuit grow ~5-fold in length and receive ~5-fold more synapses, but maintain a constant postsynaptic density. Axons grew less than dendrites, but they compensate by increasing their presynaptic density by 30%:

3: A new study uses existing connectome data from the fly ventral nerve cord to better understand how flies walk.

They built firing-rate models where the activity of all 4,604 neurons evolved based on the summed inputs from connected neurons, which were weighted by actual synapse counts. They randomly assigned biophysical parameters sampled across 1,024 simulations. When they simulated the activation of descending neuron DNg100, rhythmic motor patterns emerged at ~7-15 Hz, which matches fly stepping frequencies.

They next systematically removed neurons from the simulations while checking if the rhythms persisted. This allowed them to isolate just a 3 neuron circuit (E1, E2, and I1), present in the area controlling each leg, as sufficient for generating walking rhythms. They then validated this finding by optogenetically activating DNg100 in headless flies, which walked at speeds matching their simulations, as well as DNb08, which produced the predicted rhythmic leg movements.

This is an example of how doing simulations of relatively simple neuron models on public connectome data can reveal functional neural circuits. There is a massive potential research area here IMO.

4: A expansion microscopy study demonstrates the precise positioning of molecular labels in neuronal sub-components:

5: Study finds that the remodeling of mitochondria near dendritic spines could be a way to fuel the energetic demands associated with synaptic plasticity. A nice example of using electron microscopy to elucidate how the brain functions.

6: Virtual reality exploration of 3D brain structures based on electron microscopy as an educational tool. Very cool but also kind of eerie IMO.

7: A South Korean cohort study of 28,020 patients who had received cataract surgery found that those receiving blue light-filtering intraocular lenses had a 16% higher relative risk of developing depression compared to those receiving clear intraocular lenses (27.8 vs 24.1 per 1000 person-years). This is a tantalizing finding that chronic blue light deprivation through filtering intraocular lenses may impact mood. It adds to the debate about blue light-filtering intraocular lenses, which have been speculated to lead to better retinal protection, but on the downside could increase the risk of falls. It would be nice to see stronger data before putting too much stock into this finding.

8: The company eGenesis received FDA clearance for a clinical trial of a genetically engineered pig kidney for transplant patients with end-stage kidney disease. In their ongoing expanded access study, the first patient has maintained kidney function for over seven months (the longest-surviving recipient of a pig organ), and a second patient successfully received a transplant in June 2025.

A question to ponder: which would cause more benefit: spending more time and money to accelerate this type of xenotransplantation research, or donating a kidney oneself?

9: Meta-analysis of studies with 391,043 total participants finds that in people with ADHD treated with stimulants, the risk of developing a psychotic disorder (2.29%) or bipolar disorder (3.72%) is non-negligible. Like other studies, it finds that this is more likely with the use of amphetamines as opposed to methylphenidate (OR: 1.57).

10: The Rumpelstiltskin effect: the potential healing power of a diagnosis.

11: A controversial argument that olanzapine and risperidone may be close to as good as clozapine for schizophrenia.

12: Pathological laughter as an extremely rare side effect of the SSRI sertraline.

13: There is no strong evidence of increased suicidality in people who use GLP1s, finds a meta analysis of randomized control trials.

14: A phase 2b randomized trial of single dose MM120 (a “pharmaceutically optimized” form of LSD) for generalized anxiety disorder (GAD) in 198 adults finds that doses of 100 µg and 200 µg reduced Hamilton Anxiety Rating Scale scores by 5.0 and 6.0 points respectively versus placebo at week 4, while 25 µg and 50 µg doses did not show a significant benefit.

15: A randomized trial of 99 patients with idiopathic normal-pressure hydrocephalus (NPH) found that shunt surgery significantly improved gait velocity and balance at 3 months compared to placebo. However, cognitive and bladder symptoms showed no significant improvement. The shunt group also experienced more subdural bleeding (12% vs 2%) and positional headaches (59% vs 28%) compared to the placebo group. Here is a short post of mine from 2017 with skepticism about the diagnosis of NPH.

16: Study finds that in the postmortem brains from people with schizophrenia, there is a selective 18% reduction in NMDA receptor density at synapses targeting parvalbumin-positive interneurons. This is consistent with the NMDA hypofunction hypothesis of schizophrenia. Not a huge effect, but intriguing.

17: A new study suggests that wild barley harvesting was present at the Toda Cave (in modern-day Uzbekistan) 9,200 years ago, with ancient barley grains found alongside sickle blades and grinding stones. This expands the known range of pre-agricultural cereal foraging beyond the Fertile Crescent, showing that plant harvesting with specialized tools was performed across a much wider area of Eurasia than previously thought.

18: Do you know what causes brain rot? (Aside from Twitter and TikTok.) Dying.

A study develops a new measure to quantify the extent of brain decomposition in forensic cases:

Here’s what the different stages look like:

And they found that the stage clearly correlates with the postmortem interval:

19: An argument that to get to a trillion dollars, biotech companies will need to (successfully) target age-related diseases.

20: An obituary for Ricardo Insausti, a neuroscientist and anatomist in Spain who did transformative research on the hippocampus and brain banking methods, and who died in late 2024 after having cancer for 2 years. I believe that his aldehyde perfusion-based brain banking method, which was published in 2023 and was the culmination of decades of work, is currently the most well-validated whole brain preservation protocol ever published.

I was lucky enough to speak with him twice, once over Zoom when I was working on a perfusion fixation review and once when he came to visit my mentor’s lab in New York and we went for lunch. Both times he was extremely kind, open minded, and encouraging. Reading his obituary, it is clear that he was a great man who made many invaluable contributions to the world.

21: This is one of the most interesting Manifold markets I’ve ever seen:

The idea is that a company has created an oral supplement (Evolv GLP-1) using a genetically engineered yeast strain that expresses a proprietary peptide called the “EV1 Peptide.” This peptide is supposed “mimic” the effects of both GLP-1 and GIP hormones (like tirzepatide), by being absorbed into the body and binding to both of these receptors to reduce appetite. The product is direct to consumer as a supplement, and, with a subscription, it costs $126 for 4 weeks (so $1632/year).

Does it actually work? Highly unclear to me. Am I planning to take this myself? No. Would I recommend for others to take it? Also no. It hasn’t gone through the safety testing that I would consider standard for a medication.

But is it interesting? Of course it is. It’s an oral weight loss supplement that anyone can buy online that might actually work fairly well. If it works (and is allowed), it could disrupt a huge industry.

There was one part of the company’s website that didn’t make any sense to me, and I found concerning:

What does it mean to be “specifically formulated to have no harsh side effects”? I guess this could be dancing around the words “designed” and “formulated”, but I don’t consider it particularly forthright. The Titanic, after all, was designed to not sink.

I don’t see how it is possible for the supplement to be effective and also have no possibility of harsh side effects. They argue that the fact that it is shorter acting would reduce side effects, which might be true, but it wouldn’t entirely eliminate them. I don’t see why some of the GLP-1 side effects, like gastroparesis, could not still be seen with the use of Evolv.

Another interesting facet is the legal situation. Basically, in the US, companies can make their own determination that a substance is Generally Recognized As Safe (GRAS). They could argue that the base organism (yeast) is GRAS, the genetic modifications are as well, and the produced compounds are similar to naturally occurring substances.

The FDA could decide that they are not having the “GRAS” designation. The FDA has previously taken action against supplements making medical claims or containing pharmaceutical-like compounds.

If the GRAS designation is allowed by the FDA, and their delivery system is as good as they say it is, then this could allow other bioactive peptides to also go direct to consumer via genetically engineered yeast supplement form. This could become an enormous market and is quite possibly the real story here.

22: Biotech company Until (formerly Cradle) reports that they closed a $58M Series A round and have raised $100M+ to date. They seem to have pivoted to reversible organ cryopreservation to aid in organ transplants, at least in the near term. It seems that facilitating organ transplants is the way that most people think it might be possible to make significant money in the field of cryopreservation.

23: Profile on the Longevity Investors Conference, a conference for high-net-worth individuals in Gstaad, Switzerland. The article focuses on Tomorrow Biostasis, a company doing cryopreservation of whole bodies. Emil Kendziorra reports that they have almost £175m in contracts from people who have signed up with them.

24: A philosophical argument that if there is a reasonable probability that cryonics (or brain preservation) works, and individual people choose to pursue it, then most of their expected well being lies in the future, not the present. They said it, not me!

It will not surprise you to hear that I think that they are wrong. The whole point of brain preservation is that while we are not able to reverse the process of long-term preservation today (trust me, if we could, you would know), we can still attempt to preserve the information in the brain so that if powerful technologies do arrive in the future (which many expect), people in the future might be able to use those technologies to revive people preserved today. An actual argument against this being technically possible requires an argument that (a) the important information cannot be preserved by current techniques and/or (b) such future technologies cannot or will not ever be developed. I welcome any such arguments.

But there is also a related question: Is research towards reviving a preserved mouse the best way to help make brain preservation successful, so that people today have the best chance of being revived in the future? In other words, should we pour most of our resources into trying to cryopreserve a mouse and rewarm it?

My answer to that is also no.

I will try to explain why. But first, I need to give you a bunch of context.

What do other people in the field think about this topic (and others)?

We recently posted a new preprint, in which we collected and analyzed predictions about the future of biostasis/brain preservation from 22 people in the field, who are mostly researchers.

Participants were gathered based on asking the speakers at Vitalist Bay Biostasis Week and their professional networks. So we should expect that they are not exactly pessimistic about the field. Otherwise, they probably wouldn’t be so involved in it.

I will use a few of the results from this preprint to frame the discussion. Of course, my opinions in this post may be different than others in the field, including the other co-authors. This is good to keep in mind, in no small part because I might be wrong. Please consider reading the full preprint.

What are the main methods used in brain preservation?

There are two major types of methods used for preservation today: (a) aldehyde preservation and (b) pure cryopreservation (not using aldehydes).

Aldehyde preservation is the gold standard for structural preservation of the whole brain, such as in immunostaining or connectomics research. On the other hand, pure cryopreservation is the gold standard in IVF, tissue banking, and organ banking for transplant.

(Very infrequently, pure cryopreservation methods are used for preserving large pieces of brain tissue for microscopy studies, for example when researchers also need to do some sort of molecular assay that is currently incompatible with aldehyde preservation. And also very infrequently, aldehyde preservation can be used in tissue banking, such as for bioprosthetic heart valves. But both of these use cases are much less common.)

Aldehyde preservation could not be used as the initial step in a preservation procedure without the use of very advanced future technology for revival. On the other hand, some believe that a cryopreservation procedure theoretically could reach this level at some point in the future, potentially requiring a relatively simple rewarming procedure.

To put my cards on the table, I’m partial to aldehyde preservation, because of the validated structural preservation that this method provides.

How far away are we from being able to do provably reversible cryopreservation of whole mammalian organisms?

Opinions on this vary widely. In our preprint, we asked participants the timeframe by which they thought it would be possible to reversibly cryopreserve a whole laboratory animal and recover it to a long-lived healthy state1. On the y-axis, you can see the cumulative percentage of respondents who estimated that this milestone would be achieved by that time frame or earlier:

Participants, by necessity, will make these timeline estimates while having wildly different worldviews on other topics. For example, based on their other responses, some of the participants seem to expect with high probability what is effectively a near-term technological singularity. Among this subset, it’s reasonable to see why their timelines would be accelerated. But most respondents don’t believe that this milestone is likely to occur until the 2040-2050 time frame, and most don’t put high confidence on it until 2070-2100.

I myself find even this distribution of estimates to be sooner than I expect. The problem just seems really difficult to me. First, a whole body cryopreservation procedure would need to be able to cryopreserve every tissue and organ in the mouse body with a single mix of cryoprotectants and rewarm them all to a viable state. But for many of these tissues and organs, we can’t yet reliably cryopreserve and warm them in a viable state even individually, with the cryopreservation procedure optimized for just that tissue or organ. Second, such a procedure would also need to deal with the fact that some parts of the body are vascularized poorly or not at all, and therefore cryoprotectants cannot be easily distributed to them by perfusion. Seems hard. There’s a reason people have been trying this for decades without much success.

Here are the predictions for when people participants thought it would possible to reversibly cryopreserve a human patient with recovery to a long-lived healthy state:

These timeline estimates are considerably further away. They are still faster than I’d expect. But again I’m not factoring in truly transformative AI, which might be an Everest regression — “controlling for altitude, Everest is room temperature” — when it comes to predictions about the late 21st century.

What do people in the field think about other future revival technologies?

Even though provably reversible cryopreservation and rewarming appears to be decades away, that doesn’t mean that the field of brain preservation is necessarily hopeless. The whole idea is that there could be an entirely different technology tree that could eventually allow for the revival of preserved people without requiring the preservation to be provably reversible with rewarming possible at the time of preservation.

What might those future technologies be? Well, we obviously don’t know for sure. In our preprint, we focus on the three classes of methods that have been most commonly discussed in the field: (1) in situ repair, such as Robert Freitas’s proposal of conventional cell repair, (2) molecular nanotechnology (MNT)-based reconstruction, such as Ralph Merkle’s proposal of off-board repair and (3) whole brain emulation, such as Ken Hayworth’s proposal of slicing a preserved brain, scanning the structures with electron microscopy, and building an emulation based on that.

Ignoring all of the other ways that preservation could not work, we asked participants what their probability estimates were that each of these three proposed revival technologies (conventional cell repair, molecular nanotechnology, and whole brain emulation) would ever be technically able to revive any people who were preserved in 2025 and earlier. Here were their probability estimates:

Whole brain emulation and molecular nanotechnology are each estimated at around a 50-60% probability of technical feasibility, and conventional cell repair at around a 25% probability of technical feasibility.

Of course, it’s hard to bring up whole brain emulation in this context without some people immediately pointing out that they don’t consider it to be revival. Which is understandable. This is a common viewpoint and it makes sense that people have strong feelings about it. But it matters, because whole brain emulation was rated in our survey as both the revival method most likely to be developed first and the most technically feasible method for people preserved today. To better understand where our participants stood on this question, we asked them whether they believe that whole brain emulation could allow for genuine continuity of experience:

As you can see, although most said yes, a significant minority in our survey thinks that whole brain emulation could not constitute genuine revival. This percentage is larger in other surveys, such as Max Marty’s survey of cryonicists, which found that around 50% of people would consider it to be survival. (Of course, even if people think whole brain emulation could constitute survival and allow for continuity of subjective experience, they still might not necessarily want it, for other reasons. These questions can and should be separated.)

Of the two main preservation methods, aldehyde preservation is sometimes strictly associated with whole brain emulation, meaning that people claim that it could only work with that method of revival. This is, in my view, a highly unfortunate association, because (a) aldehyde preservation is the gold standard for structural preservation in neuroscience research, which means it makes sense to focus on it if you think (like I do) that provably reversible cryopreservation and rewarming is very unlikely to be developed for humans anytime in the next few decades, and (b) aldehyde preservation really does seem to be compatible with molecular nanotechnology-based reconstruction, which is probably why key proponents of molecular nanotechnology, such as Eric Drexler, Robert Freitas, and Ralph Merkle, have said or implied as much. It seems to me that the molecular crosslinks formed by aldehydes could be reversed in the same ways that molecular damage from ischemia or cryoprotectant toxicity would need to be for molecular nanotechnology to ever be able to revive people preserved today.

So the next question we asked participants is whether they thought that preservation methods that use aldehydes would be compatible with molecular nanotechnology, if such technology were ever developed. The options were “Very likely”, “Likely”, “Unsure”, “Unlikely”, or “Very unlikely”. Here’s how they answered:

As you can see, nearly all participants thought that it was likely that molecular nanotechnology, if ever developed, would be compatible with a type of aldehyde-based preservation. And they also thought that molecular nanotechnology was no more likely to be compatible with pure cryopreservation approaches than with aldehyde-based ones. Perhaps you think I am belaboring this point, but the memeplex has been so contaminated on this question that I feel I must make this point crystal clear, or else some people will inevitably respond to this post by saying that they think that aldehyde preservation is a bad idea, because they don’t want whole brain emulation.

Here are a few of my predictions for the future

So far a lot of this post is me arguing against other positions. So to be more straightforward, here is my current thinking about what I do think is likely to happen with the field of brain preservation in the future, if we humans don’t blow ourselves up first.

First, I don’t think that scaling up provably reversible cryopreservation and rewarming methods that work for tiny animals or small organs is the method that is most likely to lead to revival for most individuals. Especially not for those who are legally die and choose to be preserved in the upcoming few decades.

Instead, I think we will need to rely on the development of powerful future technologies, like molecular nanotechnology or whole brain emulation (I consider in situ repair methods like conventional cell repair less likely to be feasible, because of their inability to perform detailed molecular repairs, although I wouldn’t rule it out).

How long will it take to develop these future technologies, if they are ever developed? I was part of a team that did a different survey of a few hundred neuroscientists about memory. One of the questions was about when they estimated that whole brain emulation would be possible for different species, if ever. The median estimate of these neuroscientists was that human whole brain emulation would be achieved in 2125 — i.e., 100 years from now:

This seems like a reasonable estimate to me. Right now one of the main bottlenecks for brain emulation efforts is proofreading of electron microscopy data, but I think that advances in computation will render this much faster in the upcoming 5-10 years. Instead, it seems like the main bottleneck is going to be understanding how the brain works on a mechanistic level. This is a hard problem. It depends crucially on the level of cellular detail and complexity that the brain uses for different types of information processing, which is unknown. Progress here will be accelerated to the extent that understanding how the brain works can be turned into a computational problem of how to predict electrophysiology from molecular and cellular data. It will be slowed to the extent that it requires a ton of wet lab work.

Like the majority of the participants in our survey, I think that if any revival technology is ever developed, whole brain emulation will likely be the first. But as we have discussed, some people do not and will not desire whole brain emulation, and these preferences should of course be strictly respected. I think that molecular nanotechnology will most likely be the next method to be developed, which could be used to revive preserved people who do not desire whole brain emulation. By the time that whole brain emulation is developed, the pace of technology development seems likely to be very fast. As a result, my guess is that even molecular nanotechnology approaches that seem magical to me today, like Ralph Merkle’s off board repair, will probably be able to be developed.

To be clear, I’m not claiming that these potential methods would work perfectly for everyone who has been preserved. I think that revival would still only be possible with either of these methods to the extent that the information in the person’s brain is actually preserved well enough.

If you agree with my prediction that this is likely to happen in the future, what should you do today? Well, I think one reasonable approach is to mostly ignore this, perhaps aside from signing up for brain preservation services, and do something else entirely. There are a bunch of other important problems in the world as well.

Personally, for whatever reason, I have long had enough conviction about these predictions for the future that I am really motivated to work on this problem. It seems to me like the most useful thing for me to do is to focus my efforts on structural brain preservation, to attempt to provide the best preservation of the information in the brain that we can perform today, to increase people’s chances of revival in the future.

I also want to point out that even though I think structural brain preservation is the most urgent area where work is needed, I still think that provably reversible cryopreservation and rewarming is an interesting research topic. There are a lot of synergies between these areas. Researchers in both fields use similar methods, such as the perfusion of chemicals through the vascular system. Provably reversible cryopreservation experiments in laboratory animals could also test our theories about how memories work, which could help us learn which structures we need to preserve in order to retain long-term memories. Finally, this research might help get more people interested in and excited about the field.

Does it make sense to work on developing future revival technologies today?

Some have asked: shouldn’t we be directly working on making these revival technologies? I disagree with that too. At least beyond basic sketches of how they might work.

To me, researching the application of these potential revival technologies today seems like trying to write a video game before the development of the transistor. It’s just not possible to even get meaningfully started without access to powerful underlying technologies that we don’t have yet. It’s also not possible to make a difference in the rate of development of these fundamental technologies, which are being pursued independently for other reasons that are more immediately economically valuable, unless you have billions of dollars. If someone in the 1930s could imagine the development of video games and really wanted to play them (which, I mean, fair), their best bet would have been to focus on their personal longevity and try to survive 40 or so years. Trying to make Duke Nukem with vacuum tubes would not be a good use of their time.

What exactly do I think we should be working on, then? Well, there are plenty of problems with our current options for structural brain preservation. In one of the other parts of the questionnaire, we asked participants about how likely they thought that different possible failure modes would lead to their information-theoretic death, if for some reason they were to legally die and needed to be preserved within the next month2. Here are the probabilities that they predicted for each of these potential failure modes:

The three consensus most likely failure modes were inadequate preservation quality even under ideal conditions, geographic barriers preventing timely enough preservation, and poor procedural execution.

To me, most of these seem like worthwhile problems to work on. For example, while I think that the methods we have available today are pretty good, they are far from ideal. No brain preservation method has yet been demonstrated to preserve the entire connectome in a human brain. It would be fantastic to get our demonstrated ultrastructural preservation quality across the entire brain to the next level of validation. A next step would be to do so very reliably, and under less-than-ideal conditions.

I also think that most of the options in this field are too expensive. When I was thinking about this whole topic a few years ago, I really didn’t like that I couldn’t afford to sign up myself, my family, and my close friends who wanted to do it as well. Which to me is one of the points of this whole thing… wanting to spend more time with the people I love. Also, historically people have sometimes lost money near the very end of their lives and therefore not been able to afford preservation at all. This even happened recently to a former president of Alcor! I don’t want to have to stress out about this possibility my whole life, especially when something unpredictable and outside of my control like medical bills could cause bankruptcy.

I think that if it was done locally, aldehyde-based brain preservation could be done for thousands of dollars — and ideally be subsidized by philanthropy for those who can’t afford it — rather than tens or hundreds of thousands of dollars. There’s a reason that people in the US can donate their brains to a brain bank for free. Everything involved is relatively cheap, and there is enough philanthropic funding of the topic by wealthy and powerful people who care about it to cover these costs. A lower cost would make brain preservation more accessible and maybe even help with some of the social stigma around it (it won’t solve the social stigma entirely, but that’s another discussion for another time).

Third, I think it is unfortunate that the gold standard method for preserving the structure of the whole brain, perfusion-based aldehyde preservation, is currently only offered by one organization in the world, which is Sparks Brain Preservation (and which, to disclose an obviously major conflict of interest of mine here, is where I work). It would clearly be better if there were other such options as well, for example for people living far away from us. For this, we need the field of structural brain preservation to grow. I am grateful that Tomorrow Biostasis is open to this method and performs aldehyde preservation in certain cases with high ischemia, and that Hiber and Nectome are planning to start offering aldehyde preservation services in the future. Our own organization has ambitious plans for growth. We also plan to make our methods publicly accessible for other organizations to adopt if they are interested in doing so.

All of these are practical areas where I believe we can make progress in the next few years, if there is sufficient effort and funding.

Anyway, thanks for coming to my Ted talk. Whether you’re skeptical or supportive, I invite you to consider also reading the full preprint. And if you’re interested in helping, please reach out to me.

It has 3008 cells and ~368,000 total synapses. The connectome is sort of an input-to-output sensory system, because it starts with the synaptic terminals of photoreceptors and ends with the beginning segments of 599 ganglion cell axons that form part of the optic nerve. (The actual optic nerve has around 1.2 million axons.)

The donated eyes were from a 52 year old man, who was a brain dead organ donor at the Medical University of Vienna. They removed his eyes and made a tiny cut in them to allow the fixative solution — 4% glutaraldehyde — to reach the retina. The eyes were then shipped in fixative at room temperature to the University of Washington where the retina was dissected, stained, and embedded for electron microscopy. The sample was imaged with an Apreo SEM Volumescope and the data can be viewed at a new website called NeuroMaps.

Here is a graphic showing their process for segmentation and classification:

One of their interesting findings was in the wiring relevant to color vision. They found that S-cones (a type of photoreceptor which detect blue light) make direct contacts with L/M cones (which detect red/green light) through their axon end branches, which are called telodendria. This doesn’t happen in other primates, where these cone types are isolated from each other.

They also found that H1 horizontal cells — i.e., the interneurons that create contrast detection through lateral inhibition — receive input from all three cone types in humans. This creates a trichromatic receptive field. In other primates, H1 cells specifically avoid S-cone connections and only process red/green information. This might help to explain some of the distinctive aspects of human color perception.

Another finding is that over 95% of the human foveal ganglion cells they imaged belong to just three types (midget, parasol, and small bistratified). Compared to humans, there is a wider diversity of ganglion cell types in non-human primates, and especially in mice.

This is an example of how the human central nervous system is specialized for certain tasks — in this case, precise visual acuity as opposed to motion detection and more diverse types of visual processing — but not necessarily more complex in any meaningful sense than that of other animals. In fact, in this case I think it is simpler.

A few technical notes. They used neural networks trained on mouse visual cortex data to segment the tissue, suggesting that AI methods in connectomics can transfer across species and brain regions. However, they still had to make over 37,000 manual proofreading edits to refine the segmentation. Human annotation is still a major, major bottleneck for connectomics. Finally, many of the co-authors are from the company Zetta AI, who seem to have helped with the computational aspects of the work.

2: A new preprint presents the first unified connectome of an adult fruit fly’s entire central nervous system. The flies were initially preserved via immersion in cold fixative, 2.5% formaldehyde and 2.5% glutaraldehyde. They mapped ~160,000 neurons and ~218 million synaptic connections across both the brain and ventral nerve cord (analogous to the vertebrate spinal cord).

You know this project is a BFD because they not only gave it an acronym — BANC, for Brain and Nerve Cord dataset — but they also kindly noted how that acronym should be pronounced (it’s “bank”).

One of their findings was that motor and endocrine control of the body of flies operates through local feedback loops, where the effector cells in each body part are primarily influenced by sensory neurons from that same body part, creating tight sensory-motor circuits throughout the body. These local circuits are then coordinated across the entire organism by ascending and descending neurons that link the brain and ventral nerve cord, allowing different body parts to work together in organized behavioral patterns.

Based on this analysis, they also found brain regions involved in learning and navigation (i.e. the mushroom body and central complex) act as supervisors rather than direct controllers of behavior. They claim that this is similar to the robotics paradigm where semi-autonomous modules can override each other based on behavioral priorities.

3: Study does 3D electron microscopy of the macaque visual cortex, focusing on the synaptic architecture across different cortical layers. The animals were preserved via transcardial perfusion with 4% formaldehyde and 0.125% glutaraldehyde and the embedded blocks were imaged with FIB-SEM. One finding was that the thalamic input layers (4Cα and 4Cβ) contain mitochondria-rich multisynaptic boutons that are up to 12 times larger than typical cortical synapses and likely play a role in high-fidelity sensory signal transmission.

4: Ascending venules are the blood vessels that carry deoxygenated blood outward from brain tissue back to the surface. But most of them only drain the superficial cortical layers.

In a new study in mice, researchers discovered that just 3% of these ascending venules, which are called the principal cortical venules, extend deep enough to serve as the venous drainage system for the deepest cortical layers and white matter.

They also found that during aging, the capillaries that feed into these rare principal cortical venules constrict and regress, leading to a ~20% reduction in blood flow specifically in the deep brain tissues, while the superficial layers maintain or even increase their flow. This decreased blood flow was correlated with demyelination in deep tissues. And experimentally inducing similar blood flow reductions in young mice reproduced the same type of white matter damage.

This suggests improving flow in these deep venous drainage pathways could be a therapeutic target for preserving white matter health in aging. This could be relevant to postmortem perfusion as well.

5: A neuroscientist figures out how to visualize volumetric EM data on their office wall. Pretty cool IMHO.

6: In a study of a new method for expansion microscopy, human brain tissue was autopsied from the brain of a 61 year old Hispanic man at the San Diego Medical Examiner’s Office. It was manually sliced into 1 cm coronal slabs and flash frozen with liquid nitrogen. A slab from the occipital pole was then drop-fixed in 4% formaldehyde for 12 hours at refrigerator temperature. They then cleared the fixed tissue, expanded it 4x, and labeled it with an antibody (SMI-32) that stains heavy chain neurofilaments. After all of this treatment, they reported that they were still able to clearly see axons and their trajectories throughout the whole sample under the microscope.

7: Study develops a method of perfusion fixation and laser capture microdissection followed by RNA sequencing. It’s particularly useful because previous studies have claimed that fixation makes RNA sequencing difficult or impossible. This study adds to the database suggesting that this perception is just due to the limitations of our currently available methods.

8: Study using in situ MRI scans of postmortem brains reveal that diffusion processes continue in brain tissue after death, with measurable pseudo-perfusion persisting rather than dropping to zero as one might naively expect. Reminds me of the dead salmon fMRI paper. My guess is that this is likely due to ongoing autolysis (tissue breakdown) in the tissue, since these signals correlate with the temperature and postmortem interval.

9: It has been previously found that with aging, there is a lower correlation between transcript abundance and protein abundance in cells. A new study suggests that this is due to altered translation — in particular, ribosome stalling.

10: A widely covered study finds that treatment with lithium orotate (a form of lithium that avoids being trapped by amyloid plaques) can prevent or reverse memory loss and brain pathology in mice.

In fairness, I haven’t looked into this all that much, but I’m somewhat skeptical. I did a journal club presentation a few years ago on the possibility of low dose lithium for the prevention of dementia. My gestalt takeaway from putting together that journal club is that the field seems biased to me. I would bet against lithium orotate being useful as a treatment to prevent dementia in humans. Still seems worthy of further investigation, though.

11: Study finds that the serotonin transporter SERT acts as an immune checkpoint in CD8+ T cells (like PD-1 or CTLA-4). SERT apparently suppresses their anti-tumor activity by decreasing the local levels of serotonin. Common SSRIs like fluoxetine were found to suppress tumor growth in multiple mouse cancer models by blocking SERT.

Like most papers I read, I’m a bit skeptical of this story so far, since it’s mostly mouse studies, no prospective clinical trials, etc. But it’s certainly interesting and adds another wrinkle to the discussion about SSRIs and longevity.

12: In a randomized control trial, simvastatin does not improve depression outcomes in people with depression and obesity when added to escitalopram. It did improve their cardiovascular risk profile. Like many other people, I would like to see the results of this same trial design, but with an anti-obesity medication instead of a statin.

13: Thomas Reilly covers a fascinating study finding that pre-treatment with CBD does not prevent (and actually worsens) the psychosis-causing effects of THC in people with schizophrenia. This is counter to previous suggestions that CBD may help prevent the pro-psychosis effects of THC. Quite impressive to get this study actually done, and useful data.

14: Another trial finds that naltrexone treatment mitigates the acute antidepressant effects of ketamine, once again suggesting that the mechanism of action of ketamine is highly related to allosteric modulation of the opioid system. It’s consistently surprising to me that people don’t really seem to care about this.

15: Trial of saffron extract (n = 165) for insomnia finds that there was a significant reduction in insomnia scores, when combining the saffron groups (30 mg and 20 mg) versus placebo (P = .035), although the effect size was not large (d = 0.45). Tantalizing but certainly not a slam dunk.

16: Study abstract notes that sexual minority (e.g. gay, lesbian, or bisexual) people are 9x more likely to receive a diagnosis of OCD.

17: Steven Byrnes has some interesting ideas on how the brain works. He distinguishes the “steering brain” (i.e. the brainstem/hypothalamus with hardcoded business logic) and the “learning from scratch brain” (i.e. the cortex/cerebellum/striatum). He speculates that morality might just be “another cell group next to the one that controls your blood pressure” in the hypothalamus. And that it’s important to figure this out so that we can make sure that powerful AI systems of the future are safe even if (and in his view, when) they rapidly become extremely intelligent.

18: I want to signal boost Sacha Gusev’s comment on embryo selection:

19: Abi Olvera notes that Montana has now passed a right to try law. Fantastic news in my opinion.

20: Study finds that long-distance runners are much more likely to have advanced adenomas on colonoscopy. This may be related to the previous finding that long-distance running causes ischemic bowel injury, potentially due to reduced splanchnic blood flow in intense exercise.

21: Interesting job posting from Ed Boyden on LinkedIn, hiring for a “Stealth mode Neuro AI research startup.” Job description: “Research scientist to analyze existing, and in-the-pipeline, ground-truth oriented (i.e., showing fundamental building blocks and their interactions) neuroanatomical and molecular brain map data, and physiological and causal data, from multiple species (e.g., fly, human, mouse, etc.), to extract principles of brain computation, including network motifs, patterns, and equations.”

22: Vitalik Buterin believes that humans will need to merge with AI, first through BCIs and eventually through mind uploading, to maintain agency once AI gets much better in the future. He thinks this will be necessary to avoid becoming “cows grazing on the fields” while AIs do all the important work. He sees uploading as the best of three non-extinction options (the others being human irrelevance or an attempt at a permanent AI pause), that would allow humans to think at the “technological ceiling” and remain productive contributors rather than just being passive beneficiaries of AI progress.

23: For a few years now I have felt that Alex Wissner-Gross is one of the most underrated thinkers in the world. Here’s a recent podcast with him. He discusses the need for AI benchmarks on important open scientific problems that will help human welfare (“abundance benchmarks”), especially ones that can be mechanically verified. He argues that AI alignment is a civilizational task requiring coordination across researchers, corporations, and governments, not just individual efforts. On human-AI integration, like Vitalik he also predicts that many humans will choose to merge with machines as a way to maintain relevance in the future age of highly advanced AI capabilities.

A common experimental paradigm to study memories is contextual fear conditioning. Specifically, laboratory mice are trained to associate a spatial context or a tone with an aversive foot shock. After successful learning of this association, mice will freeze when they are exposed to the tone (orange) or the spatial context (blue):

A new study uses a bunch of neuroscience tools to show that changes in the input synapses to a particular brain region involved in fear memory — the basolateral amygdala — are responsible for this contextual fear conditioning.

In one of their experiments, they implanted a new associative memory into mice. To do so, they first used a retrograde synaptic virus to introduce a channelrhodopsin into neurons in both the basolateral amygdala (BLA) and their presynaptic partner neurons in different regions across the brain:

They next exposed the mice to a spatial context (Box B) and a foot shock on separate days (to avoid a natural memory), while using the stabilizing drug trimethoprim lactate (TMP) 20 minutes after exposure to stabilize the expression of channelrhodopsin in the neurons that were activated in response to those stimuli. This allowed them to specifically express channelrhodopsin in the neurons responding to the presentation of both of these stimuli:

Then, by optogenetic stimulation of the mice that had channelrhodpsin in the neurons responsible for encoding memories about both Box B and the foot shock (shown in the black line), they were able to artificially create a fear memory specific to Box B. This resulted in freezing behavior when mice were placed into Box B:

Other experiments in the study show that the fear memory formed leads to changes in the ratio of AMPA receptors to NMDA receptors at input synapses to the basolateral amygdala. The authors summarize their findings:

Engram neurons are active during learning, reactivated during retrieval, and essential for memory recall. Our study further demonstrates that synaptic plasticity underlies these properties, as [basolateral amygdala] neurons activated during learning and expressing immediate early genes: (1) undergo specific synaptic changes at learning, (2) are reactivated during retrieval, and (3) require these synaptic modifications for memory recall.

The study doesn’t tell us exactly what the structural changes are at synapses. It also doesn’t do decoding of memories based on a static map of the preserved mouse brain after learning, which theoretically should be possible if their results are true. Although these results are novel in several ways, this type of study has been done before, which is part of why this particular study probably hasn’t make a huge splash since it is was published. But to me that actually makes it more interesting, because it’s a more reliable result.

Overall, I think they present good evidence that synaptic changes are necessary and sufficient for this type of fear memory. And because it’s hard to see why other types of memory would work differently, it’s another piece of evidence that other long-term memories in the brain are also formed in this manner.

2: New study presents a correlated light and electron microscopy data set of the brain and anterior spinal cord of a 7 d old zebrafish larvae:

With this data set, they have so far produced an incomplete connectome, with approximately 20% of the total neurite path length assigned to neuronal cell bodies. Significant amounts of axon tracing would be required to extract more connectome information. It was still enough to draw some new insights about how the zebrafish larva brain works.

To preserve the brain before doing the connectome mapping, they anesthetized the animal, dissected it to expose the brain, and then fixed it by immersing it in 2% glutaraldehyde and 6% mannitol alongside microwave irradiation.

They found that in the spinal cord, the quality of tissue preservation decreased alongside the direction of the tail, which led to a decline in the segmentation quality in that direction. Never having preserved a zebrafish brain myself, nor having read much about the field, I wonder if adding formaldehyde to the glutaraldehyde solution might have decreased this preservation artifact, because formaldehyde penetrates much faster than glutaraldehyde and the gradient suggests to me that it might be a problem of penetration speed.

3: A few months ago, we covered a study led by Alexander German in which the authors demonstrated functional recovery of adult mouse brain slices after vitrification. Now the preprint has been updated to also include similar functional recovery after perfusion-based vitrification of the whole mouse brain in situ, making the paper an even bigger advance, as it is the first study in the world to ever report this.

I apologize for a bunch of technical details here. I find it fascinating, but obviously skip if you find it boring.

First, they tried performing transaortic perfusion with gradually increasing concentration steps of cryoprotective agents. When they did this, the brains became so dehydrated, to 44.5 ± 1.8% the volume of control brains, that they could not measure any evoked field postsynaptic potentials (figure i below). Next, they performed “shock-loading” via immediately perfusing their full-strength vitrification solution (called V3), which made the dehydration not quite so severe, to 55.7 ± 3.0% the volume of control brains, but still too dehydrated.

They then performed a “interleaved equilibration” approach in which they alternated the perfusion of full-strength vitrification solution and LM5 to rehydrate the brain. This led to vitrification with normal brain volumes (iii below), but caused edema when they subsequently washed out the cryoprotective agents (iv below). Their final perfusion solution therefore used a partial interleaved equilibration approach and partial dehydration, with vitrified brains at 69.8 ± 13.5% the volume of control brains (ii below). The brains were stored at -140°C for 1-8 days, and the cryoprotective agent was washed out at low flow rates and with a solution that has a high osmotic concentration.

They found that this final perfusion protocol worked in one out of three iterations. The other brains had to be excluded, as they had inadequate perfusion quality or undetectable electrophysiological activity. They then studied electrophysiology of hippocampal slices from the successfully perfused brains. They found some differences compared to control brain slices, but overall the electrophysiology was highly similar:

It’s certainly interesting to speculate why only one out of three perfused brains led to good electrophysiological results. But given the pace of this group’s research so far, I wouldn’t be surprised if they solve the problem of unreliable perfusion in the near future.

4: A nice example of immunogold labeling of postmortem human brain tissue, in this case with an antibody for phosphorylated tau. The tissue has a postmortem interval of a few hours and it was immersion fixed prior to processing for electron microscopy:

5: In research studies, blood tests seem to be getting increasingly good at predicting dementia years in advance. In one study, GFAP levels in the blood were found to start rising 10 years before the symptoms of Alzheimer’s disease were seen.

6: A method for sectioning brain tissue that creates a custom mold for each brain based on a segmentation of an MRI scan for that brain. I thought this was a nice example of using 3d printers.

7: This month in “GLP1s treat everything”: a study finds that liraglutide reduces monthly migraine episodes by half, and a retrospective study finds that GLP1s used in patients with ovarian cancer have a survival benefit (HR of 0.45) that would be among the most significant outcomes in the extant literature on ovarian cancer treatment. However, while these medications are obviously amazing, we do need to be cautious of GLP1 hype. I suspect that the ovarian cancer finding is confounded in a way that is hard to control for, because patients with more severe disease would likely have more weight loss as a result of their cancer, thus preventing the GLP1s from being prescribed.

8: João Pedro de Magalhães on problems with senolytics as an anti-aging therapy. Doesn’t seem to be going very well. Senolytics were considered very promising a few years ago.

9: New study finds that a potential anti-aging drug, rapamycin, is not effective for Alzheimer’s disease, in an n = 10 trial. Alzheimer’s disease biomarkers worsened. They also found that rapamycin does not penetrate into the CSF.

10: Study finds that plasmapheresis (i.e. removing or replacing aged blood plasma) is not an effective anti-aging therapy in humans, and instead seems to accelerate epigenetic aging clocks.

11: Meta-analysis of preclinical studies of anti-aging therapies led by Austin Parish finds that most findings in non-mammal do not translate to mammals.

12: Study of the neuropathology of centenarians finds that amyloid accumulation is associated with significantly worse cognition, especially when it happens alongside tau accumulation. This is evidence in favor of the amyloid hypothesis, although it runs into the classic problem that amyloid could be a marker of other pathological mechanisms.

13: Launch of a 5 year ARPA-H program of replacement of brain tissue as a therapeutic approach. The two focus areas are on generation of tissue and methods for engrafting it to promote recovery of brain function.

14: A new study called DEPRE'5 randomizes people (n = 257) with major depression and inadequate response to an SSRI to 5 different protocols: optimization of the SSRI dose, adding lithium, adding nortriptyline, switching to venlafaxine, or adding problem-solving therapy (30 minute sessions, at a frequency of slightly more than once a week). Over 6 weeks, they found that all of the non-SSRI dose optimization approaches outperformed, with the addition of nortriptyline having the most side effects and problem-solving therapy having the least.

15: Thomas Reilly on a new randomized study (n = 150) of pramipexole as an augmentation agent for treatment-resistant depression. It seems quite effective.

16: Post on LLMs exacerbating or causing psychosis by Tom Pollack, a neuropsychiatrist, who was part of a team that posted a preprint on this topic. This has become a hot topic. It is certainly worthy of consideration, although I always get worried about moral panics around things like this. My main comment is that I don’t think that the intersection of psychiatric disorders and people’s digital lives is going away anytime soon.

17: And here’s a nice academic article also by Pollack on how “inflammatory reductionism” is causing harm by attempting to explain many psychiatric conditions as caused by inflammation or immune dysfunction without good evidence. Co-sign. Although effects of the immune system on the brain is certainly an exciting area of research, not everything is due to inflammation. We have no choice but to attempt to tolerate ambiguity in our understanding of mental illness.

18: Trial of bright light therapy in bipolar disorder finds that up to 45 minutes is safe and relatively effective for depressive symptoms as long as there is close monitoring, with a low rate of hypomanic switches that resolved quickly with dose reduction.

19: Eric Drexler on a new software program that allows people to design molecular machines.

20: A new “AI neuroscience” company called Prima Mente. So far, they seem to be focused on building models for the diagnosis of neurodegenerative disorders based on blood tests.

21: Interview with Peter Thiel, touching on longevity, cryonics, and whole brain emulation.

Douthat: To keep us in the concrete, I want to stay with that example for a minute and ask: OK, what does it mean to say we need to take more risks in anti-aging research? Does it mean that the F.D.A. has to step back and say: Anyone who has a new treatment for Alzheimer’s can go ahead and sell it on the open market? What does risk in the medical space look like?

Thiel: Yeah, you would take a lot more risk. If you have some fatal disease, there probably are a lot more risks you can take. There are a lot more risks the researchers can take.

I remember 1999 or 2000, when we were running PayPal, one of my co-founders, Luke Nosek — he was into Alcor and cryonics and that people should freeze themselves. And we had one day where we took the whole company to a freezing party. You know a Tupperware party? People sell Tupperware policies. At a freezing party, they sell ——

Douthat: Was it just your heads? What was going to be frozen?

Thiel: You could get a full body or just a head.

Douthat: The “just the head” option was cheaper.

Thiel: It was disturbing when the dot matrix printer didn’t quite work, and so the freezing policies couldn’t be printed out.

Douthat: Technological stagnation once again, right?

Thiel: But in retrospect, it’s also a symptom of the decline, because in 1999 this was not a mainstream view, but there was still a fringe boomer view where they still believed they could live forever. And that was the last generation. So I’m always anti-boomer, but maybe there’s something we’ve lost even in this fringe boomer narcissism, where there were at least a few boomers who still believed science would cure all their diseases. No one who’s a millennial believes that anymore.

Douthat: I think there are some people, though, who believe in a different kind of immortality right now. I think part of the fascination with A.I. is connected to a specific vision of transcending limits. And I’m going to ask you about that after I ask you about politics. …

Thiel: We’re jumping around a lot of things. So, again, the critique I was saying is: They’re not ambitious enough. From a Christian point of view, these people are not ambitious enough. Then we get into this question: Well, are they ——

Douthat: But they’re not morally and spiritually ambitious enough.

Thiel: And then are they still physically ambitious enough? And are they even still really transhumanists? And this is where, OK, man, the cryonics thing that seems like a retro thing from 1999 — there isn’t that much of that going on. So they’re not transhumanists on a physical body. And then, OK, well, maybe it’s not about cryonics, maybe it’s about uploading. Which, OK, well, it’s not quite — I’d rather have my body. I don’t want just a computer program that simulates me.

Douthat: Yes, I agree.

Thiel: So that uploading seemed like a step down from cryonics. But then even that’s part of the conversation, and this is where it gets very hard to score.

We can already create computer programs which simulate the brain with some degree of accuracy, but most people don’t think that qualifies as life extension. A brain emulation — which we cannot yet do, and probably won’t be able to for a long time, if ever — would be different. It would theoretically preserve all of the relevant causal relationships and information processing of the original brain, on some sort of different substrate. In my understanding of the situation, an accurate whole brain emulation would create a branch of the person and allow them to live on in the physical world in a biological or robotic body, or in a virtual world in a virtual body. I don’t see how this is unambitious. For example, it seems to me to be obviously the best way to travel to other stars.

22: Cryopets has hired Chana Phaedra:

23: My talk from May at Vitalist Bay about our work in structural brain preservation is now online. Thanks so much to the organizers for putting this conference together.

Using fluorescence microscopy to target specific hippocampal layers, they then perform focused ion beam-based cryo-lift-out to extract tissue chunks in serial. They section these into 5 μm slices, and thin them to 150-300 nm. This enables them to collect electron tomography tilt series that reveal nanometer-scale structural details like synapses. It’s a complicated process but at the end they can visualize synapses across volumes without the use of chemical fixatives:

This is not the first study to show this, but their good synapse visualization is another example that synapses can clearly be preserved in brain tissue that is well vitrified. Here’s a 3d projection of one of the synapses they were able to see:

It’s also interesting to note that the process leads to ischemia for 1.5 to 4 hours prior to preservation via high pressure freezing, but they found that this duration of ischemia did not affect synaptic vesicle diameters.

They had some trouble finding a protocol that did not lead to any ice formation, requiring them to try out 18 different combinations of cryoprotectants. It may be useful to learn from their failures, even on such small samples with the use of high pressure freezing.

I guess that the ice formation could have been more easily prevented by using a higher concentration of cryoprotectants, but possibly using higher concentrations would have led to more dehydration, ischemia, osmotic damage, or all of these, which would have then made the synapse visualization less effective.

2: New paper uses the Drosophila connectome to provide the first comprehensive map of how touch-sensing neurons in a fly’s head connect to its brain circuits. They suggest that this explains the anatomical basis for how flies automatically groom specific body parts when they are touched there.

Note that these are innate behaviors, not learned memories. But Tomás Ryan et al. have previously argued that innate behaviors and learned memories likely have similar neural underpinnings, calling the structures underlying innate behaviors “ingrams” by analogy to the “engrams” that underlie learned memories.

3: New study measures hippocampal neurons using calcium imaging in mice while they learn reward locations over the course of a week. They found that place cells are reformed daily through behavioral timescale synaptic plasticity (BTSP), which is a phenomenon where calcium plateaus in dendrites strengthen synapses that fired seconds earlier.

Most place cells are unstable and disappear after 1-2 days. But some cells become increasingly likely to reactivate at the same location each day, though the mechanism driving this increased probability is unclear. This growing stable population of place cells correlates with better task performance over time.

What this suggests is that hippocampal place cell memories require reconstitution through plasticity occurring during each experience, though some unknown structural trace must persist to guide the cells and synaptic connections in which this plasticity is biased to occur.

It seems that there is a trade-off between plasticity and stability, allowing neural systems to maintain enough flexibility for new learning while preserving enough structure to reliably reconstruct previous spatial memories when they are needed.

Another note is that it seems to me it wouldn’t be possible to preserve the behavioral timescale synaptic plasticity in a static structure. It’s much too ephemeral. But the structural trace biasing these representations to form in particular ways seems like it could potentially be captured in a static state.

4: Neuromodulator-annotated electron microscopy reconstruction from larval zebrafish:

5: All-synthetic memory inception in the vinegar fly via optogenetic activation of both sensory and dopaminergic neurons:

6: Proteomic profiling of a brain that was excavated from a former cemetery in Bristol, UK. As far as I can tell from the article, this brain is from someone buried several decades to around 200 years ago. Given the circumstances, it was surprisingly well-preserved, with identifiable cerebral hemispheres and intact sulci and gyri. They took a biopsy sample of it from the anterior right frontal lobe.

They developed a protocol to measure the proteome of the brain, using urea lysis. Their data suggests that that hydrophobic proteins preferentially survive decomposition, perhaps because they precipitate and aggregate in the acidic conditions of decaying tissue. On the other hand, their extraction methods recovered mainly acidic peptides. These peptides especially came from cytoskeletal proteins, which seemed to be relatively well-preserved over the long term.

7: We can learn some details about the operations of brain banks from this article. For example, regarding the Mount Sinai Brain Bank:

Unsure whether it’d be able to pay staff after April, Mount Sinai Brain Bank sent layoff notices to its nearly two dozen employees, then revoked them when it learned of the funding extension less than 18 hours before the deadline, said program director Harry Haroutunian.

In the month before the contract expiration, the program, which stores 2,640 brains in 43 freezers set at minus 80 degrees Celsius, stopped taking [new brain donations]. Mount Sinai typically takes two brain donations a week.

One takeaway here is that brain banks require a large number of staff and funding to bank brains from new donors, which in this case is provided by the federal government. But long-term storage of the already banked brains, especially the ones in fluid preservation, is a much less resource-intensive process.

8: Profile of neuroscientist Davi Bock at the University of Vermont. Discusses connectomics and its implications. Interesting ending:

As I head out of the building, I think about going back in to ask him more about uploading our brains. In our conversation, he said that, even if we could, he wouldn’t want to, afraid that the digital copy of his brain would be low-fidelity. “It’s very likely to be some weird, altered thing that just doesn’t have much to do with me,” he said. I want to ask him if he’d like to make a copy of his brain if he knew that it would be a perfect copy—but then I think the better of it and keep walking out into the sunshine. His reasoning seems clear, and he’s made up his mind.

9: New study does exome sequencing on 500 people with obesity. Finds that “Monogenic obesity was diagnosed in 5.8% of patients, while 7.1% carried a potentially obesogenic variant” and “Surprisingly, diagnostic yield was lower in severe obesity cases. 40% of patients with monogenic obesity carried variants in genes not included in current obesity panels.” The most common mutation they detected was MC4R, followed by PHIP and SRRM2. Here are a few examples:

10: New method for clot removal does spinning to densify the clot’s fibrin network and allow red blood cells to be released.

11: In older mice, facial massage is found to help get lymphatics flowing and restore CSF outflow to youthful levels. 🧐

12: Review of potential treatments for intellectual disability. Most success in the space so far has come from early treatment of metabolic disorders. They argue that different treatments may be possible for a broader set of disorders.

13: New GWAS for sleep duration. The strongest association they found was with rs2863957 on chromosome 2, which is within the PAX8 gene, explaining ±2.4 minutes in sleep duration. Together, the variants explain ± 220.5 minutes of sleep duration across all genome-wide significant loci. The variants are most associated with genes known to function in lung, liver, immune, brain, and eye cell types, in that order. Although I kind of don’t totally trust this type of enrichment analysis, to the extent that it is true, this is an example of the importance of whole body function for what I would personally think of more as a brain phenotype.

14: Paradromics does its first testing of its brain computer interface implant in humans, just for 10 minutes in tissue that was going to be removed for epilepsy surgery regardless. They’re using a device with 420 electrodes that record from individual neurons.

15: Comparison of the contrast achieved by different heavy metal stains in electron microscopy.

16: I created a Manifold market on when there will be a full mouse connectome. Not many predictions yet! Criteria:

Requires synapse and cellular connectivity mapping across one entire brain (i.e. not combining data across multiple specimens). Does not require a particular technology, e.g. could be electron microscopy or expansion microscopy based, but does require synapse-level resolution and accurate tracing of cell processes. Some degree of potential inaccuracy or very small gaps is accepted, to the same level as the adult Drosophila connectome, e.g. as described here: https://www.nature.com/articles/s41586-024-07968-y. Resolves upon publication or preprint or public announcement of completion, does not require peer reviewed publication to be accepted.

17: Aschwin de Wolf on how funeral directors can assist with brain preservation/cryonics. I am most interested in the extent to which embalming protocols could be adapted to perform aldehyde perfusion for structural brain preservation.

18: New 5 million euro funding round by Tomorrow Biostasis.

19: Interview with Laura Deming about preservation and personal identity.

A lot of good material here. First, I totally agree with this take: “I think because cryo is so weird, it’s not worked on. There's also actually a bunch of other reasons, but there’s a lot of baggage around it that makes it an idea that has like a force field around it. But then once you’re inside it you’re like: this is so beautiful and so impactful. It’s so underworked on for what it could be.” She goes on to call it “anti-memetic,” which I also agree with.

Second, she has some interesting thoughts on personal identity (some editing for clarity and brevity, feel free to check out the video/transcript):

Interviewer: So say I reject the ego view of personal identity and I accept Derek Parfit’s bundle theory [that] each of us is just a web of experiences shifting through time. How should that affect how I think about longevity?

Deming: I'm very obsessed with this question right now… I feel like not enough people are thinking about it even though some people are thinking about it… I think the question is, what do you want to preserve? And you can actually make arguments for continuity, you can reconstruct reasons that you might care about physical continuity over time. But I think they’re very different from what you might be born with, which is kind of this feeling of to survive you need to make it to the next moment. I think that becomes hard to defend unless you just want to pick it intuitively as a thing to hold on to, which you can also make the choice to do that.

Interviewer: How close are we to the point at which these debates start influencing how capital is allocated?

Deming: Today.

Interviewer: Seriously?

Deming: 100%.

Interviewer: Can you say more about that?

Deming: I think that today they’re influencing how capital flows in a very subconscious way. Most people are born with certain beliefs around these topics that are not that well examined. And they’re guiding intuitions about what is correct and what is not correct to invest in today. This is interesting because the investments today then determine what might be most available in the future and that might determine what people have access to. …

Interviewer: Which view do you think is winning at the moment, at least in Silicon Valley?

Deming: What I see personally, although it might be very biased, there’s a lot of starting out with what we’re born to be most adapted to, which is I just want my physical self to continue for as long as possible. And just the most conservative possible perspective on preserving yourself. Like Ship of Theseus kind of a way.

I think when people really think about it often there’s like kind of a one-way door — or maybe it's a two-way door — but a door you go through where it's like oh actually like it’s pretty hard to defend that. In Buddhism this might be like analogous to doing like a no self meditation, just repeatedly asking “what is the I,” “what is the I,” “what is the I.” I'm not that knowledgeable, but I think in Buddhism when people just do this enough, eventually they’re kind of like well actually there's like — maybe they come closer to the Derek Parfit view, which is interesting that both types of philosophical traditions kind of get to a similar place. But then I think you should still be doing things to keep your body healthy and alive and so you kind of have to reconstruct notions of what’s meaningful about that.

I agree that people tend to start out with — “are born with” — views on personal identity that require maintenance of the physical continuity of one’s atoms over time. But that with further examination, they often come to views on personal identity such as branching identity, and if they do so, they tend to stick with those.

20: Another perspective on preservation and personal identity by a new Substack, “This Body is Beta.” Links to a couple of my favorite articles on the topic. The author on the goals of the newsletter:

“If you’re curious about the real science of aging, what it means to preserve identity, excited about breakthrough technologies that both have an immediate impact on health right now and advance longevity research…or simply love someone mortal, “This Body Is Beta” is for you 🖤

Last year I thought I was dying. This is my love letter to staying alive 💌”

21: Article by Aurelia Song and Charlie Dever in Asterisk Magazine about brain preservation. I agree with many of the points made and I disagree with a few others. Disagreements are more interesting, so I will focus on those.

One of my disagreements is regarding the 12 minute window they posit is required for high-quality ultrastructural preservation, due in turn to a requirement for high-quality perfusion. What is nice is that this is an empirical question, so the disagreement can theoretically be settled via observation (either with new data or references to existing data in the literature).

For now what I will offer is a couple of questions. First, how do we know that this window is the same in humans as it is in mice and rats? It seems to me that it could be different, especially if the cause of death is different (euthanasia vs respiratory depression).

Second, how do we know that there is no biological variability across brains in the extent of perfusion possible after a given period of ischemia? For example, in their study on the no-reflow phenomenon in cat brains, Ames et al 1968 found that after the same 10 minutes of ischemia, 0% of the brain of rabbit #33 was white and non-perfused, while 23% of the brain of rabbit #48 was. My guess is that, rather than a threshold, there is more likely a spectrum in how perfusable a brain is after a given period of ischemia.

I’m also curious about this:

I immediately notified Alcor myself, even though he had been dead for two hours and counting. Something is better than nothing, right? In the end, it took multiple days for Alcor to recover him, and he was straight-frozen.

Around that time, as part of my research, I happened to be working with a human brain preserved via immersion fixation after a similar time delay. I wanted to see for myself how well the synapses were preserved, so I took samples from the outermost, best preserved tissue in that brain, which was likely to be an upper-bound for the quality of my mentor's preserved brain.

Like Darwin's early work in cats, the preliminary results were encouraging: Using an ordinary light microscope, I could see all of the cells and nerve bundles, and there was no obvious physical damage. But the important information in the brain is really in the microstructures between cells — the synapses. And these tiny details are too small to see under a light microscope.

So I prepared a sample for an electron micrograph. What I saw devastated me. The brain was shredded. Not only were the synapses ruined, the entire structure was riddled with holes. That was what my professor's brain must look like.

When we did electron microscopy on the cortex of immersion fixed human brains after around 1 day after death, we did not identify any “holes.” Here’s a link to the raw data. And I believe that our findings are consistent with the literature, as discussed in the paper. However, we might be missing something.

It’s possible they are referring to what we have called ambiguous interstitial zones, which are probably due to a local expansion of extracellular space. These are quite concerning, but not necessarily indicative of complete information loss — in my view, it’s still unknown. I’m interested to hear more details.

Most importantly, I appreciate that Aurelia is pushing for higher standards in this field, which I agree is urgently needed.

22: New interview with musician Sufjan Stevens. I agree with him that there’s a greater power in survival than in strength. And also that sometimes survival requires being open to transformation.

You know, strength is funny because it suggests a kind of power and authority and vigor. But I also think there's greater power in survival. And sometimes survival requires sensitivity and openness, and even subservience. I think I've just become a lot more subordinate to the chaos of the world around me and less inclined to fight it, because I'm starting to learn that you cannot create change by force: You just have to move through it, and be open to transformation.

Michał Januszewski has been at Google Research in Zürich working on connectomics since 2015. He was the lead author on a preprint posted this month that proposes a new software method called PATHFINDER. This method is estimated to dramatically reduce the need for manual proofreading, an 84x decrease from the optimistic estimates of the Wellcome Trust report. That would save… a lot of money.

(This publication was obviously a team effort, but I also want to note this from the author contributions section: “MJ developed the PATHFINDER system, applied it to the EM volume, and evaluated and visualized the results.”)

How did they do it? They combined three AI models: (a) a Flood Filling Network that creates supervoxels with very low merge errors, (b) a new model called SENSE that proposes agglomeration candidates, and (c) another new model called SHAPE that evaluates whether or not the proposed neuron shapes are morphologically plausible.

Each model operates at a different spatial scale, progressively building up from small “supervoxels” to complete neurons. The Flood Filling Network works on a tiny 0.5 µm field of view to create high-quality base segments. SENSE then operates at 2-4 µm to identify which segments might belong together. Finally, SHAPE looks at 20 µm regions to evaluate whether proposed combinations actually look like real neurons. Here are some visualizations of how the model works on the preserved mouse cortical tissue that they tested it on:

They focused their analysis on the ability of PATHFINDER to trace axons, which generally have a smaller diameter and greater length than dendrites, making them more difficult to trace. Comparing the segmentations to manual ground truth annotations, they identified two major types of errors: (a) split errors, which is when a single axon is incorrectly broken into multiple separate segments, and (b) merge errors, which is when distinct axons are incorrectly fused together into one segment. Naturally, there is a trade-off between split and merge errors.

They then defined a metric of success, which is the normalized expected run length (nERL). The expected run length measures the average axon length correctly traced from a random starting point, with any segment containing a merge error scoring zero to penalize this type of error. This was normalized to the maximum possible error-free path length in the ground truth annotations. Here’s what they found:

As you can see, the normalized expected run length is much better with PATHFINDER than alternative algorithms. They estimate that this accuracy would dramatically decrease the need for manual proofreading, making the number of proofreaders needed only an order of magnitude higher than was needed for the Drosophila connectome.

When I initially read the Wellcome Trust report, I was naturally bullish, because I like many others have noticed the rapid increase in AI capabilities in recent years, and the proofreading problem seemed solvable with advances in AI. But let’s be clear. “AI” didn’t do this research project. Michał Januszewski and colleagues did. Commendation is due to them for their efforts.

I do have one quibble. It’s not really a critique of this paper in particular, but moreso the whole field, and might simply reflect my own ignorance (although I have tried to research this for awhile). For context, here’s Supplementary Figure 1:

When I look at these types of 2d images, for the most part I cannot clearly tell the difference between axons, dendrites, and astrocytes. I recognize that this might just be a skill issue, and that, although the images shown are in 2d, to actually distinguish structures for the segmentation, annotators have access to the 3d data and need to use it.

However, here is what I would like to know: what is the inter-rater reliability for these ground truth annotations? How sure are we that each of the green-colored structures are axons as opposed to dendrites, or microglia, oligodendrocyte, or astrocyte processes? 95% sure? 99%? 99.9%? I didn’t see any such interrater reliability data for the “ground truth” human annotation, either in this paper or in Shapson-Coe et al 2024. Perhaps this would require something like correlative light and electron microscopy data set to be very sure, but I would at least be reassured by seeing high overlap scores on a subset of the human annotated data.

2: Researchers use optogenetic techniques to study the role of the posterior dorsomedial striatum (pDMSt), which is a brain region that receives visual input. They trained mice to associate an optogenetic cue (light activation of neurons in the visual cortex) with a food reward obtained through forelimb reaching. Surprisingly, while optogenetic inhibition of the posterior dorsomedial striatum completely blocked new learning over weeks of training, it had no effect on recalling already learned associations, even those formed just hours earlier. The authors note that this aspect of memory storage and recall likely occurs elsewhere in the brain. I found these results interesting because I find that the distinction between learning and memory recall is highly under discussed.

3: New study suggests how memories transform from precise, episodic recollections to generalized, “gist” memories over the first few weeks after formation. In particular, they found that newly formed memories activate specific neuronal ensembles only in the original context, but these same neurons become “promiscuously active” (🤔) in similar but distinct contexts over time. This enables generalization. They also find that this phenomenon is dependent on neurogenesis. It’s interesting to think about this process of generalization in the context of psychodynamic psychotherapy.

4: Which network structures best allow the prediction of a neuron’s functional activity? New study analyzes this using matched electrophysiology and connectome data from n = 497 neurons in the MICrONS dataset. They first embed the neurons into a 2D hyperbolic space based on their connectivity patterns, i.e. which neurons they share inputs and outputs with. They chose a hyperbolic space for dimensionality reduction because their analysis and previous studies have suggested that this better matches the brain’s hierarchical, tree-like organization better than flat Euclidean space.

They use a leave-one-out method to predict the electrophysiological activity of a given neuron based on the activity of all of the others, weighted by the similarity of its neural connectivity to those other neurons:

They found that the position of neurons in this 2d hyperbolic embedding space predicts neuronal activity with 68% accuracy, which was surprisingly higher than the 56% accuracy achieved when using the full connectome. This suggests that neurons with similar “connectivity fingerprints” tend to have similar neuronal activity patterns. My guess is that readers of Olaf Sporns’s book “Networks of the Brain” will appreciate this paper.

5: New study uses EEG and measures respiration during two visual categorization tasks in n = 25 humans. They find that respiration seems to modulate brain activity and the encoding of external signals 200-300 milliseconds before behavioral responses. In other words, EEG can help measure how we perceive the world differently based on our respiratory phase. This is a good example of how cognition is embodied.

6: New blog by Ed Boyden, Claire Wang, and Nina Khera. The first post is on what classes one should take to become a neuroengineer. I don’t see what the point of two courses on quantum physics is, but I could be missing something.

7: A study of over 600,000 women veterans found that the prescription of an antidepressant medication (SSRIs, SNRIs, and other types) was associated with a significantly lower risk of developing cardiovascular diseases such as heart disease and stroke over nearly 9 years of follow-up (relative risk: 0.27-0.76).

8: 23andMe has been bought by Regeneron for $256 million, who plans to continue its personal genomics service. Seems like good news for the world, since most of the value from such human genomics data sets is the aggregated scale + phenotype data. Although Regeneron is a reputable company and this seems like a good outcome for 23andMe customers, the whole saga does show a key downside with relying on a for-profit company to survive and remain aligned with the interest of customers over the long term.

9: A study on the renewal of NIH grants finds that a lack of funding interruption is associated with less switching of research topics and more novel findings. The obvious bias is researcher quality, but the study authors did their best to adjust for this. The results are consistent with the idea that scientific innovations often require more than 10 to 20 years of continuous work.

10: In a 24 month follow up of a pilot study (n = 18), wearing a weighted vest for ~10 hrs/day during a 6 month period of diet with calorie restriction seems to help with decreasing weight regain afterwards. This is consistent with a gravitostat model of how the body regulates the resting metabolic rate (RMR). However, I find the idea that the weighted vest provides long-term benefits even when it is no longer used pretty unlikely, so I won’t be convinced until it is replicated with a larger sample size and more statistically reliable differences.

11: Comprehensive study of mixed pathology in post-mortem human brains, with a focus on the medial temporal lobe. They study Alzheimer’s disease neuropathological changes (ADNC), microinfarcts, small vessel disease (SVD), α-synuclein, TDP-43, granulovacuolar degeneration (GVD), argyrophilic grain disease (AGD), and others:

First, they found that many but not all of these pathologies are correlated with one another:

Second, they found that lower neuron counts in the CA1 region of the hippocampus are associated with a higher risk of cognitive impairment:

Third, they found that many of the pathologies were associated with a decreased neuronal density, lower brain weight, and/or clinical dementia rating (CDR):

One basic take-away is that there are number of different types of pathology that the brain can accumulate that have been associated with cognitive impairment. And these are just the relatively common and well-known ways. And we cannot meaningfully reverse the vast majority of them.

12: Study on the genetic correlates of treatment resistant depression in n = 292, 663 participants. Finds that there is a positive association with insomnia (OR = 1.11, 95% CI 1.07-1.15) and specific neuroticism (OR = 1.11, 95% CI 1.07-1.16), and an inverse association with educational attainment (OR = 0.88, 95% CI 0.84-0.91), but they are small effects.

13: Study on cognitive impairment after CAR T cell therapy in mice finds that a key mechanism is an increased reactivity of white matter microglia and a resulting decrease in the number of oligodendrocytes. Also finds that microglial depletion rescues the cognitive dysfunction, even if started a few weeks after CAR T.

14: Article on laws and regulations related to brain banking in Italy. We learn that in Italy, legal death pronouncement requires either (a) “observation of consecutive abiotic signs, over a period of 15- and 30-h post-mortem” or (b) demonstration of a flat EKG for 20 consecutive minutes, which is called “thanatography”. The situation in Italy for brain banking seems to be complex. The authors note the importance of developing brain donation-specific regulations, because it deteriorates much faster than most of the rest of the body, so it is critical that the brain banking procedures can be performed rapidly.

15: New results on dielectric warming of kidneys and ovaries cryopreserved via vitrification, by Brian Wowk and colleagues. Reports that they reversibly preserved three rabbit kidneys in this manner and that one survived long-term with return to normal function in a recipient that was still alive 17 months after. They noted that this kidney was 13.9 g, which is the largest vitrified vital organ that has ever been shown to return to normal clinical function, by an order of magnitude.

16: Ariel Zeleznikow-Johnston has a summary of the Vitalist Bay Biostasis Week, which I attended and found to be fantastic. Thanks so much to the organizers for putting it on. I will add to Ariel’s comments that Ralph Merkle and Max More both had excellent talks and comments on the talks of others. To me the most encouraging thing was to see the wonderful new entrants in the field over the past several years — you know who you are! I would personally love it if more people joined the field of structural preservation in the coming years. I am still learning myself, but I will offer to help mentor you if you are a researcher or communicator looking to join or switch to this field, even if part-time.

But when I read his Wikipedia page (a website which one must read with caution) I learned that he was also convicted of bribery and corruption by parliament, imprisoned, and removed from his position as Lord Chancellor of England. What was up with that? I kind of didn’t know what to think about it.

On the one hand, Lisa Jardine and Alan Stewart’s “Hostage to Fortune: The Troubled Life of Francis Bacon 1561-1626” was sort of a typical biography, insofar as it chronicled Bacon’s life from birth to death. But I think it is better thought of as basically a long contextualization and explanation of exactly what happened during his long political rise and rapid downfall, and how someone so brilliant could find himself on the wrong end of court intrigue. And after reading it, it turns out that Bacon’s ideas are even more interesting once you know about Bacon the man.

⭑

In this review, I am going to try to make a long story short. Bacon was a prototypical courtier, meaning that he was attempting to be a companion or adviser to the king or queen. He was also a lawyer and a philosopher, but mainly he was a courtier. The game for such people at the time seems to have been to make the monarch happy with you — and thus give you good jobs and other benefits — without making everyone else too annoyed. This required establishing and managing a complex network of alliances, favors, and information.

Bacon had several advantages in this game. First, he seems to have been smarter than most of his contemporaries. There were one or two situations described in the book where he was able to cut right through to the core of a confusing matter in a way that the other people in the room were not.

Also — and this is where Bacon has been criticized over the years — he seems to have been willing to … not let principles or friendships get in the way of his success. For example, he initially hitched his fortune to that of Robert Devereux, the Earl of Essex, which was quite helpful to his position for a time. He then abandoned Essex just prior to his downfall and participated in the legal proceedings against him, joining the prosecution team at Essex’s treason trial. When the tide turned back in favor of Essex after James I — whom Essex had supported — became king, he wrote his “Apologie in Certaine Imputations Concerning the Late Earle of Essex,” defending his actions. Jardine and Stewart seem to cast some shade on Bacon for this type of behavior, but this was the game at the time, and it’s hard to blame the guy for being good at the game.

It doesn’t help his case that despite Bacon’s success, he still managed to go into tremendous debt, because he spent so much money on a massive retinue and splendid parties. (This may have contributed to his fall, because he may have been more likely to accept gifts that he shouldn’t have.)

Naturally, his rise also involved competition with others who wanted the same positions as he did. One such person was another giant of the time, Edward Coke. Coke took the first round in 1594, when he was appointed Attorney General by Queen Elizabeth, thanks to his partnership with the powerful Cecil family. Bacon took the second round, when in 1613, he managed to trap Coke into insisting to King James’s face that the monarch was not entitled to special consideration in litigation. As a result, Coke lost his position and was almost imprisoned, while Bacon was appointed Attorney General. It seemed that Coke was out of the picture.

But Coke’s story was not done yet. First, Coke orchestrated a controversial marriage between his 15-year-old daughter Frances and John Villiers, brother of the Duke of Buckingham (the King’s “favorite”), despite his wife’s objections and the girl’s resistance. Bacon tried to prevent the marriage, which seems from my vantage point to have been in good faith, as Frances did not consent, Villiers seems to have had serious mental illness, and they separated a few years later. But it was a huge tactical error, as the King strongly favored the match, and it helped Coke regain royal favor while diminishing Bacon’s standing.

Then, when the King called Parliament in 1621 to raise revenues, Coke positioned himself as a leading member of the opposition. Here, Coke got his chance for revenge, as he helped a parliamentary committee to charge Bacon with 28 counts of corruption for taking gifts or bribes from people whose cases he was judging. At the time, bribe-taking was common practice and it was not necessarily considered corrupt if it didn’t influence judgment. This seems to have been the case for Bacon, as, ironically, the people who testified against him seem to have been mostly annoyed that the bribes didn’t work. Bacon argued that it was a situation of vitia temporis — abuses of the times, rather than vitia hominis — culpable behavior of a person beyond this. But the Parliament wasn’t having it.

Bacon saw the writing on the wall, confessed, and escaped with his life. However, he was fined £40,000 (perhaps around $20-27 million today), imprisoned briefly in the Tower of London, barred from future office, and prohibited from entering the “verge” — i.e. the 12-mile area around the court. This was a dramatic fall, engineered by his bitter lifelong rival.

⭑

I am sorely tempted to frame the next part of Bacon’s life, from 1621-1626, as a story of resilience and perhaps even reinvention. And to a certain extent, I think there is a case for this. When Bacon got banished, he retreated to history (a great love of his) and read about the lives of others statesmen who had lost significant authority and turned to writing. He found three authors who he considered “the most eminent and the most resembling” of his situation — Demosthenes, Seneca, and Cicero. He noted that “the remembrance of their calamity is now as to posterity but as a little picture of night-work, remaining amongst the fair and excellent tables of their acts and works.” That’s what he wanted for himself. So he examined how they “did bear their fortunes, and principally how did they employ their times.”

Einstein of course had his annus mirabilis in 1905 when he published four groundbreaking papers in the journal Annalen der Physik. Not quite at the same level, but this period of life for Bacon — ages 60-65 — has been called his “quinquennium” (sidenote: quinquennium is a great word meaning five years and should be used more). It is when he published much of his famous philosophical work. Yes, he had started a lot of this work over the years and had clearly been thinking about it. But my guess is he probably never would have gotten around to writing The historie of life and death, Sylva sylvarum (or, A natural history in ten centuries), or New Atlantis without this five year period to stop focusing on politics and the accumulation of money and status for once in his life.

The cheerful take-away might be that no matter your age, it’s wise to always have an intellectual project in your back pocket — ideally a great book or two — in case you get banished beyond the verge. This would allow you to at least pursue intellectual endeavors, to buoy the loss of other forms of status (career, financial, reputation, etc).

But actually, he seems to have been somewhat miserable during this period, despite his immense intellectual productivity. He certainly said as much in his letters at the time to the King and his former friend the Duke of Buckingham (e.g. writing “I am now full three years in misery”).

Actually, it seems to be a testament to his intellectual prowess that he was able to be so despondent about his situation on the one hand, then turn on a dime and become immediately productive. For example as his editor Tenison wrote about him:

[One] day he had sent a friend to court to receive for him a final answer touching the effect of a grant which had been made him by King James. He had hitherto only hope of it, and hope deferred; and he was desirous to know the event of the matter, and to be freed, one way or other, from the suspense of his thoughts. His friend returning told him plainly that he must thenceforth despair of that grant, how much soever his fortunes needed it. ‘Be it so,’ said his Lordship; and then he dismissed his friend very cheerfully, with thankful acknowledgments of his service. His friend being gone, he came straightway to Dr Rawley, and said thus unto him: ‘Well, Sir, yon business won’t go on; let us go on with this, for this is in our power.’ And then he dictated to him afresh for some hours, without the least hesitancy of speech or discernible interruption of thought.

I guess this is the real skill that philosophers and scientists have required throughout the ages, very much into today — how to deal with the slings and arrows (and grant rejections), and still actually get work done.

⭑

Some of the reviewers online have argued that this book includes too many original quotes. I completely disagree. In my view, that is easily one of the best parts about the book.

First, Bacon had a brilliant way with words. He could turn a phrase that resonates across centuries. There’s a reason for the fringe theory that he might have contributed to Shakespeare’s plays.

Second, it was lovely to read all of the Latin phrases, which were usefully translated in parentheses after, and appear to have been peppered throughout written texts of that era. (The same is true of The Anatomy of Melancholy, but because there were no translations in my version of that, it was unreadable.) One of my favorites was “Dies diem docet” — the day instructs the day. Used in context, it means that knowing the future is impossible, so we will have to figure out solutions to any problems that arise in the future on a day-by-day basis.

However, I do think that the book meandered. For example, the amount of content about his brother Anthony Bacon in the first third of the book was clearly excessive. A little bit would have been okay, and my apologies to Anthony, but we came here to read about Francis. My guess is that the authors fell into that most common of traps — including content because it took a long time to research, rather than because it was interesting and relevant. They needed to kill more of their darlings. Possibly having two co-authors made this worse, because it must have turned the editorial process political — indeed, mirroring the content of the book! Honestly, this was a huge problem, the book could have been way shorter, and as a result I can’t really recommend it.

⭑

Another reason I was drawn to this book is the title: “Hostage to Fortune.” Merriam-Webster defines this as: “something (such as a promise or an action) that someone has made or done that may cause problems in the future.” It is derived from Bacon, who wrote in a 1625 essay that “he that hath wife and children hath given hostages to fortune.” It’s hard to disagree with him. When we bring children into this world, we place those we love most deeply into the unpredictable hands of fate. This could undoubtedly cause problems.

Recently I have been thinking even more than usual about fortune, parenting, antinatalism, the future of humanity, my wasted youth, etc. And of course, I have also been thinking about brain preservation.

A few weeks ago I came across Jianfei Shao’s article “Cryonics Wager: An Analysis,” which argues that cryonics is not necessarily a good idea because people (or AIs?) in the future might not treat you well upon reviving you. Basically, there’s a real risk that even if cryonics works, life upon revival could be so bad that you’d rather just have died instead. Many others have argued this over the years. Tyler Cowen, Julia Galef, and Luke Muehlhauser, to name a few.

In fact, this fear of the future, of making oneself a hostage to fortune for even longer, is in my experience the number one most common underlying reason that people are not interested in pursuing cryonics/brain preservation. And fair enough.

Reading about Bacon’s life is certainly a reminder that one’s fortune can change rapidly and unpredictably. There are no guarantees. And yet for Bacon, even after his fall, he was still obsessed with longevity. Indeed, he seems to have died in an ill-fated attempt to extend his own life further. Perhaps for some people, life is worth the risks, scary as they might be, because it is just way too much fun, and posterity much too interesting.