NREMmodFC

Wilson-Cowan implementation for the testing of NA and ACh NREM-FC modulation hypotheses.

https://www.biorxiv.org/content/10.1101/2025.02.06.636812v1.full : Noradrenaline and Acetylcholine shape Functional Connectivity organization of NREM substages: an empirical and simulation study

• Abstract:

Sleep onset is characterized by a departure from arousal, and can be separated into well-differentiated stages: NREM (which encompasses three substages: N1, N2 and N3) and REM (Rapid Eye Movement). Awake brain dynamics are maintained by various wake-promoting mechanisms, particularly the neuromodulators Acetylcholine (ACh) and Noradrenaline (NA), whose levels naturally decrease during the transition to sleep. The combined influence of these neurotransmitters on brain connectivity during sleep remains unclear, as previous models have examined them mostly in isolation or only in deep sleep. In this study, we analyze fMRI data obtained from healthy individuals and employ a Whole Brain modeling to investigate how changes in brain neurochemistry during NREM sleep, specifically involving ACh and NA, affect the Functional Connectivity (FC) of the brain. FC analysis reveals distinct connectivity changes: a decrease in Locus Coeruleus (LC) connectivity with the cortex during N2 and N3, and a decrease in Basal Forebrain (BF) connectivity with the cortex during N3. Additionally, compared to Wakefulness (W), there is a transition to a more integrated state in N1 and a more segregated state in N3. Using a Wilson-Cowan whole brain model, informed by an empirical connectome and a heterogeneous receptivity map of neuromodulators, we explored possible mechanisms underlying these dynamics. We fit the model adjusting the coupling and input-output slope of the Whole Brain model to account for ACh and NA, respectively, and show that region-specific neurotransmitter effect is key to explain their effects on FC. This work enhances our understanding of neurotransmitters' roles in modulating sleep stages and their significant contribution to brain state transitions between different states of consciousness, both in health and disease.

• Summary: Analyzing average whole brain FC matrices of 15 individuals in W and NREM substages, we report a fluctuation in the integration and segregation profile across sleep stages,. We fitted a model to these matrices varying the local coupling and input-output slope of each node, in a homogeneous way and also using NA and ACh proxy maps. We found a general improvement in the goodness of fit using these proxys for NA and ACh down-regulation in sleep, compared to homogeneous and shuffled-map modulation modalities. We also report a positive correlation between LC-brain functional connectivity and local integration of nodes, and an analogous relationship between BF-brain FC with segregation, in the awake brain.

• Steps for reproducing the results:

1. Empirical BOLD signals were extracted using empirical/extract.py, and the corresponding timeseries are available in https://zenodo.org/records/16755776 . Empirical mean FC matrices were also calculated therein, and are available in empirical/mean_mat_*_8dic24.txt
2. Run optimize_SC_Hopf.py for optimizing the interhemispheric connections in the structural connectivity matrix. The output is SC_opti_25julio.txt
3. Sweep the G and sigma parameters homogeneously in whole_sweep_both.py to generate FC matrices, and save the gof against all states (W,N1,N2,N3). This will be the baseline against which we compare goodness of fit of the "map" and "shuffle" modality of all states. This step was run in a computing cluster, so the code is implemented to be run in slurm using multiple cores. The output has to be collapsed to a file like output/sweep_delta_homoW_fromG0.16_sigma7.68_maps_0_0_9dic24_50iter.txt
4. Find the optimal for W and use it as a baseline for what is next (can be done in heatmaps.py). Run whole_sweep_both_maps.py for the "map" modality, i.e. with the DIST_LC_proj.npy and DIST_VAChT_feobv_hc18_aghourian.npy for NA and ACh, respectively (available in empirical/maps/). This will calculate the goodness of fit against all states using the heterogeneity maps. Collapse.
5. Use whole_sweep_both_maps.py, now with the shuffled versions of the heterogeneity maps (also available in empirical/maps/), to obtain the goodness of fit of the model considering this control modality.
6. Given the optimal parameters, integration and segregation parameters per node can be obtained in run_many_seeds.py. Given the empirical FC matrices, empirical integration and segregation profiles can also be obtained here.
7. The scripts that generate the figures were used to calculate all comparisons that are reported in the manuscript.