Reviewer 1 comment 2 - Millisecond Golgi synchrony (van Welie et al)

[Helveg]

2) The authors simulated two scenarios of the cerebellar subnetwork constructed by BSB. The first is the resting state activity and the second is the impulse response of the cerebellar network. The results of their model simulations are in good agreement with previously published in vitro or in vivo physiological recordings. I’m curious whether their model could also reproduce finer scale experimental findings, such as the one by van Welie and colleagues (2016, Neuron), where they show millisecond synchrony between Golgi cells in vivo, which requires gap junctions and is enhanced by sensory stimuli.

[Helveg]

• [x] Check for millisecond synchrony between Golgi cells based on van Welie et al 2016

WelieEtAlHausser_Neuron2016_SyncGolgiGap.pdf

[Helveg]

• [x] Show electrical graph of Golgi network (https://graphviz.readthedocs.io/en/stable/manual.html)
• [x] Study sync/coupling in terms of this graph (eg clustering)

[Helveg]

• [X] start 5500ms

• [x] possibly eleminate stim if due to grc

• [x] panel C & E

  • [x] pairs within <100um distance AND directly connected
  • [x] express as z-scores

• [ ] panel C: compare to knockout

• [x] panel E:

  • [x] extra trace for knockout
  • [ ] ~extra trace for anisotropic~
  • [x] change to pair based approach & show variance

• [x] panel F: reproduce, together with the stepped index and strength index (+ pos & graph layout in suppl)

• [x] knockout gap junction: redo sims (cla pizdaint)

Predictions:

• [x] (an)isotropy: kernel shape, note in discussions, possibly ours is TOO anisotropic.

[Helveg]

• Overfitting issue: there are only a few shortest paths of 5 in the network, they happen to be more synced than the 4: we'd have to run this again with multiple reconstructions, to show that this is random, or that gocs actually are more synced again over distance (this is unintuitive)
  • [ ] ~Reconstruct network multiple times, rerun sims (unfeasible due to time)~
  • [x] Consider the last 5-path bin invalid outliers of network construction, switch to another trend test, use the outliers in the statistical analysis, but don't show their bin in the image: this way we consider them for our metrics, but don't disrupt the plot.

[Helveg]

• [x] Sim with GABA, 2.5 gap

• [x] ~Sim with 10x gap no GABA [error]~

• [X] Sim with 5x gap no GABA

• GABA disturbs temporal processing.

• x5 gap no GABA too much sync spikes in the network (80% instead of 30%)

--> Try 2.5x gap with GABA

• [ ] Test different sync measures (sync.py)

[Helveg]

εύρηκα 🎉

• gap x 2.5
  • observed millisecond precision in directly coupled cell, dissapears with KO.
  • coupling between gogli-in-the-middle cells (2 steps).
  • no coupling in 3 steps or KO
• 35% synchronized spikes. high variability in firing patterns of Golgi (is correct -> insert ref) causes high std dev in golgi pairs spike sync, but SEM is 3%
• We predict:
  • Millisecond precision control requires direct gap connections, does not propagate through Golgi cells.
  • A shape of the kernel of Golgi inhibitory control in our network.

[Helveg]

Draft panels