Action Potentials for July - by Andy McKenzie

Neurobiology Notes

SubscribeSign in

Action Potentials for July

Andy McKenzie<br>Jul 25, 2026

Share

1: Expansion microscopy, first developed only a bit over a decade ago (shameless nerd bragging alert, I wrote the Wikipedia entry on it in October 2015), has now exploded. It looks poised to allow for major advances in connectomics in the coming years.<br>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.<br>So far, the degree of 3d expansion has usually been around 16-22x (two iterative rounds of around 4.5x each).<br>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.<br>There is obviously a lot of chemistry involved that I won’t pretend to totally understand:

https://www.biorxiv.org/content/10.64898/2026.05.31.729018v2.full.pdf<br>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:

https://www.biorxiv.org/content/10.64898/2026.05.31.729018v2.full.pdf<br>And when they imaged a synthetic protein, they could identify individual peptide fragments of it in their original places:

https://www.biorxiv.org/content/10.64898/2026.05.31.729018v2.full.pdf<br>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.<br>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.<br>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.<br>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.<br>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.<br>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:

https://www.biorxiv.org/content/10.64898/2026.01.29.702630v2<br>Here’s how it works chemically:

https://www.biorxiv.org/content/10.64898/2026.01.29.702630v2<br>3: Long-distance tracing of sparsely labeled single axons in mouse brains using a new method for projectome mapping:

https://www.biorxiv.org/content/10.64898/2026.06.29.734841v1<br>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?<br>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.<br>5: A deep learning model trained on 1.7 million unlabeled EM images:

https://www.biorxiv.org/content/10.64898/2026.06.06.730367v1.full.pdf<br>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?

https://pubs.acs.org/jprobs/article/doi/10.1021/acs.jproteome.6c00200/5164102/Molecular-Solution-to-the-Paradox-of-Ancient-Brain<br>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...