Action Potentials for October
1: Our new paper on histologic brain age estimation via multiple instance learning dropped this month. Led by the brilliant Gabe Marx. The model predicts brain age within a mean absolute error of 5.45 ± 0.22 years. Most of the attention signal is in the white matter:
2: A new study analyzes the synaptic connectivity patterns of the mouse visual cortex. Nice examples of how the underlying electron microscopy data corresponds to the neural circuitry diagrams:
3: The NIH's BRAIN Initiative introduces BRAIN CONNECTS, a program aiming to develop advanced technologies for detailed brain mapping across species. There are an initial 11 grants totaling $150 million over 5 years, including the development of new tools, like ultra-high speed electron microscopy and machine learning.
4: Google announces they are planning to map the connectome of the mouse hippocampus, with expected completion in 2028:
5: A new 12-step protocol to infiltrate osmium into an entire mouse brain without cracks or fragmentation. A very impressive advance, but still not close to a human size brain. A SOTA chemopreservation protocol for whole human brain preservation at an ultrastructure level — putting aside the cost of osmium and other downsides of this for a moment — would still require slicing the brain at around 5 millimeter thickness.
6: Reconstruction of synapse structure from serial section electron tomography data. Here is an example of one reconstructed synapse broken down into parts:
7: An advance in single-molecule protein sequencing, as a new nanopore sensor and machine learning allows an accuracy of 98.6% in distinguishing all 20 proteinogenic amino acids.
8: New spatial molecular atlas allows mapping of the expression of many genes across the mouse brain:
9: The expression of clustered protocadherins allows for millions of possible combinations of isoforms on the neuron surface, effectively forming a distinct barcode. A new study using single-cell RNA expression corroborates that the variable isoforms of gamma clustered protocadherins have a very low similarity level between neocortical neurons in mice.
10: In a new study, mutation experiments of the chromatin remodeler protein cBAF show that condensate formation and protein-protein interactions are driven by the intrinsically disordered regions. A nice example of how a protein’s primary structure can predict its tertiary and quaternary structure:
11: A new study finds that the stimulation of a Gq-coupled receptor in astrocytes modulates hippocampal synaptic transmission over minutes. The authors suggest that local extracellular signaling is a potential mediator of this, since ATP/adenosine released from astrocytes has been shown to modulate the function of hippocampal synapses. I’ve become fascinated with the kinetics of neural functions since I’ve started considering the timescales criterion for determining which structures are likely to be most important for rapid neural processes, like memory recall, that can occur in hundreds of milliseconds.
12: A new study uses fMRI to investigate the neural representation of space in the human brain during navigation toward a goal. They find that in the posterior parietal cortex and prefrontal cortex, a map of the current goal is represented within a map of the current location in a geometric manner that is similar to physical space. Analogous to the place cells seen in rodents.
13: A randomized trial of MDMA assisted psychotherapy for PTSD finds that the change in CAPS-5 score was −23.7 (−26.94, −20.44) for the MDMA cohort versus −14.8 (−18.28, −11.28) for the placebo with therapy group (P < 0.001, d = 0.7).
14: In an observational Swedish cohort of patients with borderline personality disorder, use of ADHD medication was the only pharmacological treatment associated with decreased suicidal behavior (HR, 0.83), while benzodiazepines were associated with the highest risk of attempted or completed suicide (HR: 1.61).
15: Should we be getting MRI scans on everyone with first episode psychosis? A meta analysis of 1613 patients finds that 5.9% of those scanned had a clinically relevant abnormality. Most common was white matter abnormalities at 0.9%, followed by cysts at 0.5%. Still not sure how actionable these are, but it’s certainly an important question.
16: New study on the effect of insomnia on sepsis risk. A doubling of the genetically predicted insomnia risk is associated with an odds ratio of 1.37 for sepsis, only one third of which is mediated by cardiometabolic risk factors such as BMI, diabetes, and smoking. Suggests that addressing insomnia might be important in decreasing the risk of sepsis, among other aspects of one’s health.
17: In a national birth cohort study from Israel, people with a diagnosis of adult ADHD were found to have an increased risk of dementia (adjusted HR, 2.77). However, among those who received psychostimulant medication, there was no longer a significantly increased risk of dementia. Correlational, but worthy of future study.
18: Analysis of phase 3 trial data shows good results for the DORAs lemborexant and daridorexant, with a number needed to treat (NNT) for insomnia of <10 and a number needed to harm (NNH) of >10, with the most common adverse effect being somnolence. In 20 years, what percentage of people are going to be taking DORAs for insomnia and GLP-1/GIP/glucagon triple agonists for weight maintenance?
19: Sarah Constantin on how low intensity transcranial focused ultrasound works. Seems quite promising for stimulating or inhibiting targeted brain regions.
20: New study compares the color similarity judgments of humans — including those with neurotypical color vision and color-atypical vision — to judgments made by large language models like GPT-4 on color hex codes. They find that GPT-4's judgments align remarkably well with humans with normal color vision. This suggests GPT-4 accurately captures details of human color perception not matched by simpler color space models like RGB or LAB:
21: Eric Minikel on the commitment of Ionis Pharmaceuticals to perform a clinical trial of antisense oligonucleotides (ASOs) for prion disease. He notes that (a) PrP-lowering ASOs work in animal models and have by far the best chance of any current medication of working and (b) Right To Try laws are often misunderstood, as they give patients the right to request experimental drugs, but companies can still refuse access.
22: The Oslo Chronic Fatigue Consortium puts out a viewpoint article with a descriptive title: “Chronic fatigue syndromes: real illnesses that people can recover from”.
23: The findings of yet another splashy Science paper are questioned, as researchers present data showing that there are actually only three meningeal membranes in the brain. Sigh. The upside is that this is a nice example of the scientific process eventually working as intended, c.f. Julie Siegenthaler’s commentary.
24: Gwern on the heuristic that it's best to ignore "this chemical might be a subtle poison" type of research. Some small number of these findings might be true but it's too hard to know and the prior probability is quite low since there are so many of these claims that mostly turn out to be false. Also Gwern on integrating personal wiki’s with LLMs and modular brain emulation by building a brain foundation model and then specializing it to a specific subject.
25: The genetics of individuals of Europe ancestry is primarily shaped by three ancestral components: Western Hunter-Gatherer, Anatolian Neolithic, and Steppe’s Yamnaya. This population structure is generally regarded as a confounder in genome wide association studies, and scientists try to correct for it. But a new study finds that heart rate, platelet count, and many other traits are actually directly associated with these ancestral components. As with other fields, adjusting for confounders is hard.
26: More data suggesting vitrification protocols are superior to slow freezing in ovarian cortex preservation. Basically — vitrification protocols offer the same levels of molecular markers suggesting preservation quality, while having advantages in terms of time and cost effectiveness.
27: An interview with Taya Maki, President of the Société Cryonics de France. She notes that currently in France it is considered illegal to preserve a body through cryonics, but it is still legal to transport a body from France to perform cryopreservation elsewhere. Hopefully France will change this regressive law soon.