Initial E/I Analysis of V1 Ecephys Data (sub-820459)

[anoushkac08]

Hi everyone,
Following Dr. Lecoq's suggestion to start with the V1 electrophysiology recordings, here are my initial findings from exploring session sub-820459 (2025-11-12).

Methods: I filtered to 1025 SUA units and classified putative excitatory vs inhibitory neurons using a 0.4ms peak-to-valley waveform width threshold — a standard approach for Neuropixels data.

Results:
719 putative excitatory, 306 putative inhibitory neurons (E/I ratio ~2.4:1)
Inhibitory cells fire ~2x faster on average (6.8 Hz vs 3.2 Hz)
E/I firing rates are strongly correlated over the session (r = 0.918), consistent with inhibitory stabilization of the excitatory population

Currently working on the PSTH for standard vs omission trials. Happy to receive any feedback or pointers on next steps!

[jeromelecoq]

I would consider selecting E and I cells that are likely very close to one another to make sure you are looking at a local network. The probes are very long!

[anoushkac08]

Thank you for this feedback. You're absolutely right that the probe spans too much to treat as one population for local E/I dynamics, I forgot about that in my initial approach. I've redone the analysis using 200µm depth bins to isolate locally colocated E and I neurons.
The results were more interesting than I expected. local E/I correlation varies substantially with depth, ranging from r≈0.02 (not significant) at depth 1000µm to r≈0.87 at depth 400µm. The pattern isn't monotonic and there appear to be two distinct dips (around 1000-1200µm and 1800µm) separating regions of strong coupling. This could reflect layer-specific differences in local circuit function, though I don't have direct layer labels to confirm this (location field is unlabeled in this session)

I'm still working on that PSTH for standard vs omission trials next.

[jeromelecoq]

This is quite interesting actually. Can you try to do a finer binning to see if the transitions are sharper?
Neuropixels recording location are notoriously difficult to align along the shank. I wonder if this metric could help better disambiguate the transitions between brain areas. Our brain area alignment process is sometimes uncertain along the shank.
I think the dip you see at 1000 um deep is likely the white matter where many small units are likely passing by axonal events.
I would expect the balance between fast spiking and excitatory cells to potentially vary very much between cortical and sub-cortical areas.

[anoushkac08]

Hi Dr. Lecoq,

Per your suggestion I redid the binning at 50µm resolution (vs. 200µm previously) to see if the transition near 1000µm sharpens.
Result: It does, and more dramatically than expected. Local E/I correlation drops from positive values (~0.3-0.5) to strongly negative (r = -0.65, depth 1100µm) within the 950-1150µm range, before recovering to strongly positive (r > 0.3) immediately outside this zone on both sides. A smaller, similar dip appears again around 1800µm.

This aligns pretty closely with what you were saying, in that this zone corresponds to white matter. Axonal pass-through activity rather than local circuit computation would explain both the loss of correlation and its sign flip (units recorded here might reflect uncorrelated, non-local spiking activity rather than actual local E/I coupling).
If this pattern holds across other sessions, local E/I correlation could potentially serve as a complementary signal for disambiguating cortex/white matter boundaries along the probe shank, alongside existing alignment methods.
Caveat: neuron counts in the sharpest-dip bins are still modest (8-13 per group), so I'll definitely want to replicate this in additional sessions before treating it as a robust signature.

[jeromelecoq]

Yes, it would be worth seeing how it holds We are about to push out more NWB files with better areas alignment. It will be worth comparing.

[maierav]

The bimodal distribution looks promising!

How about we try digging even deeper with:
https://pmc.ncbi.nlm.nih.gov/articles/PMC10220268/

[anoushkac08]

Thank you for the suggestion! I'll take a look at WaveMAP. using just the peak-to-valley threshold is definitely a simplification, and I can see how a more nuanced clustering approach would help separate cell types that don't fit cleanly into the binary classification. I'll explore this as a next step.