Talk 19:

Next up is @acnavasolive.bsky.social talking about her impressive research on sharp wave ripple analysis across species in health and disease. Andrea is a phenomenal computational neuroscientist 🤩 and I am so excited to hear her speak for the first time!

Cont..

@acnavasolive.bsky.social discussing important considerations for identifying hippocampal ripple events in a recording session and how a convolutional NNs might help their identification.

✅

Here is our paper www.nature.com/articles/s41...

It has been a very nice journey of inter-continental collaboration, and hopefully it’s a helpful outcome for the scientific community! If you are interested in these ideas, and want to try it for your own research, please let us know how is it working for you! We are more than happy to discuss! 🤩

The manuscript has MANY more interesting analyses. For example, we saw that the intrinsic dimension of IEDs is lower than SWRs; that SWR-IED segregation can depend on the pre-processing; how is coupling between SWRs on macro- & microwires; and the unique spiking modulation patterns in SWRs vs IEDs

In addition, to speed up the curation of the detected events, we have developed ripmap, a UMAP-aided semi-automatic tool that bases its efficacy on @lmprida.bsky.social lab’s work. ripmap can effectively remove falsely labeled IEDs, reducing manual curation by up to 45%!
👉 github.com/acnavasolive...

So based on our current knowledge of rodent and human events, we started by determining what features define SWRs and IEDs, and created a pipeline to analyze epileptic intracranial EEG from start (select the optimal channel) to end (event curation)

The motivation of this work comes from the unsolved difficulties of analyzing intracranial EEG in epileptic patients, where the frequent epileptic-like events (interictal epileptiform discharges, or IEDs) interfere with SWR detection, resulting in painfully high false positives🥲

With @mojtabart.bsky.social, we applied this to human tissue, and saw that human CA3 cells appear to receive 5 times more DG inputs than mouse cells do! This finding has a lot of potential for powering up our idea of hippocampal computations.

Finally, we saw that dentate gyrus input to human CA3 (the ‘teaching’ signal) seemed very high. Luckily, we have @mojtabart.bsky.social and Hans Danzl as neighbours. Their incredible LICONN technology allows light microscopy based connectomics. See the preprint here: www.biorxiv.org/content/10.1...

Using a Hopfield-like model, @acnavasolive.bsky.social showed that expanding brain size by increasing neuronal number (sparse connectivity) is far better for memory capacity than aiming for dense connectivity (and more inputs per cell). This has some interesting philosophy for brain scaling rules!

In a fantastic collaboration with Prof Karl Rössler (MedUniWien), we applied multicell patch-clamp to human hippocampus resected from epilepsy patients. Some samples show sclerosis (disease-led cell loss), but many are perfectly intact. This is the closest to 'wildtype’ human physiology we can get..

We explored CA3, which in theory stores and retrieves memories from interconnected ensembles of pyramidal neurons. With Victor Vargas-Barroso and Rebecca Morse, we looked for these networks using octopatch. From 8 patients and 56 slices we found just 10 connected pairs! (under 1% connectivity)..

To touch base with better characterised circuits, we recorded in neocortex and saw the same dense connectivity that @yangfanpeng.bsky.social, @alleninstitute.bsky.social and others see. In fact CA3 wasn’t just sparse, but gets sparser from mice to humans - opposite scaling to neocortical circuits!

We had the first view on human hippocampal synaptic pairs, and they look slow and integrating as we would expect (perfect for associations!). However they were also much more reliable and precise than seen in rodent research.

As you may expect, human neurons were larger than mouse cells, but spine density was a lot lower, so the number of inputs from other neurons in the recurrent network doesn’t change so much. Low spine density and reliable synapses have been seen in other brain areas, so may be human circuit features

A bigger difference between human and mouse brains is the number of neurons. This has gone up by about 17 times in CA3! By (very) simple maths, our anatomy data predicts connectivity in a random recurrent network to be pretty much in line with what we record experimentally for CA3 across species.

We think this explains the connectivity and circuit scaling between brain areas - dense local circuits in neocortex, while hippocampal CA3 forms something like one big recurrent network - perfect for associating all the hippocampus’ incoming info. Circuits made to measure!