The Cerebellar Engine: Multiscale Digital Brain Co-simulations Reveal How Cerebellar Spiking Architecture Shapes Cortical Coherence

Key results

1. SuggestedRemoving mossy fiber excitation of the cerebellar nuclei (CN) lowered input-output coherence (principal trigeminal nucleus, PrV, to CN) and lowered M1-S1 gamma coherence to about the level seen with the whole cerebellum inactivated. Removing Purkinje cell (PC) inhibition of the CN raised both.

   Virtual lesions (MOStoCN, PCtoCN) in TVMB-NEST co-simulations, 10 runs per condition vs CONTROL, two-sided unpaired t-tests with FDR correction (Fig. 4A-B). Both lesions lowered the gamma-to-theta coherence ratio (Fig. 4C-D). Gamma coherence trends held without normalizing cerebellar output, but the ratio effect was then masked (Supp. Fig. S5). · 2.3.2 Cerebellar microcircuit contributions to…; 2.3.3 Cerebellar circuit elements differentially…; 4.6 Microcircuit lesion co-simulations; Figure legends

2. SuggestedAt the spike level, PC inhibition appears to decorrelate nuclear output from mossy fiber input. Synchrony between input and output fell with MOStoCN and rose with PCtoCN, while phase locking of nuclear spikes to the input gamma rhythm dropped with both lesions.

   The phase-locking value (PLV) of excitatory CN neuron (CNe) spikes to the gamma-filtered mossy fiber rate fell significantly in both lesions (Fig. 6A). Firing-rate cross-correlation and pairwise spike-train correlation between mossy fibers and CNe fell with MOStoCN and rose with PCtoCN (Fig. 6B-C). Gamma-band cross-spectral density magnitude fell with both lesions (p < 0.001, FDR-corrected; Fig. 5B). · 2.3.3 Cerebellar circuit elements differentially…; 2.3.4 Changes in cerebellar spiking activity…; 3. Discussion; Figure legends

3. SuggestedGamma power and coherence came apart. Both CN-input lesions cut CNe gamma power relative to mossy fibers, yet PCtoCN raised coherence. Lesioning molecular layer interneuron (MLI) inhibition altered PC gamma dynamics without changing M1-S1 coherence.

   Relative CNe gamma power fell significantly with both MOStoCN and PCtoCN (Fig. 3D). Relative PC gamma power fell only with MOStoCN (Fig. 3C), which the authors attribute to recurrent loops through the rest of the brain. MLItoPC lowered coherence only within the cerebellum and to a lesser extent; MLItoMLI left coherence unchanged (Table 1; Supplementary Note S2). · 2.3.1 Cerebellar microcircuit contributions to…; 2.3.2 Cerebellar microcircuit contributions to…; 2.3.3 Cerebellar circuit elements differentially…; 3. Discussion

4. ShownAfter fitting, the model reproduced the experimental drop in M1-S1 gamma coherence when the cerebellum is inactivated. This held both with neural masses alone and with the spiking cerebellum embedded.

   Gamma-band (25 to 60 Hz) coherence drop, computed as (cerebON - cerebOFF)/cerebON: mean 53.4% for TVMB-only and 54.8% for TVMB-NEST co-simulations vs 51.3% in the experimental data. cerebON vs cerebOFF was significant (two-sided t-test, p < 0.0001; 10 runs). The pathway gains were fitted to this drop using TVMB-only runs. · 2.2 The model captures experimental data across…; 4.7 Analysis and Statistics; Figure legends

5. ShownAll spiking cerebellar populations fired at physiological rates and carried whisking-related theta oscillations, although only the granule cell rate was fitted. The network was never tuned on oscillations, yet gamma power in both PCs and nuclear neurons was higher than in mossy fibers.

   The interface weight was set by grid search to a granule cell target of 4.2 Hz. Mean rates of the other populations over 10 runs fell within experimental ranges from rodent single-unit recordings during free whisking (Fig. 2F). Relative to mossy fibers, PC gamma and theta power rose; CNe gamma power rose and theta power fell (Fig. 3A-B). · 2.2 The model captures experimental data across…; 4.4 Inter-scale interface models; Figure legends