Link to Google Meet:
https://meet.google.com/gjo-tdks-ufaSalvador Dura-Bernal,
Erica Y Griffith,
Annamaria Barczak,
Monica N O’Connell,
Tammy McGinnis,
Haroon Anwar,
William W Lytton,
Peter Lakatos,
Samuel A. NeymotinWe used the NEURON simulator with NetPyNE to develop a biophysically-detailed model of the macaque auditory thalamocortical system. We simulated a cortical column with a cortical depth of 2000um and 200um diameter, containing over 12k neurons and 30M synapses. Neuron densities, laminar locations, classes, morphology and biophysics, and connectivity at the long-range, local and dendritic scale were derived from published experimental data. We used the model to investigate the mechanisms and function of neuronal oscillatory patterns observed in the auditory system in electrophysiological data recorded simultaneously from nonhuman primate primary auditory cortex (A1) and the medial geniculate body (MGB), while the awake subjects were presented with different classes of auditory stimuli, including speech.
The model A1 includes 6 cortical layers and multiple populations of neurons consisting of 4 excitatory (intratelencephalic (IT), spiny stellate (ITS), pyramidal-tract (PT), and corticothalamic (CT)), and 4 inhibitory types (somatostatin (SOM), parvalbumin (PV), vasoactive intestinal peptide (VIP), and neurogliaform (NGF)). Cells were distributed across layer 2-6, except NGF cells which were also included in L1, as these have been identified as important targets of the thalamic matrix. The A1 model was reciprocally connected to the thalamic model to mimic anatomically verified connectivity. The thalamic model included the medial geniculate body (MGB) and the thalamic reticular nucleus (TRN). MGB includes core and matrix populations of thalamocortical (TC) neurons with distinct projection patterns to different layers of A1, and thalamic interneurons (TI) projecting locally. TRN included thalamic reticular neurons (RE) primarily inhibiting MGB.
Thalamocortical neurons were driven by artificial spike generators simulating background inputs from non-modeled brain regions. Auditory stimulus related inputs were simulated using phenomenological models of the cochlear auditory nerve and the inferior colliculus (IC) that captured the main physiological transformations occurring in these regions. The output of the IC model was then used to drive the thalamocortical populations. This allowed us to provide any arbitrary sound as input to the model, including those used during our macaque in vivo experiments, thus facilitating matching model to data.
We used evolutionary algorithms to tune the network to generate experimentally-constrained firing rates for each of the 42 neural populations. We tuned 12 high-level connectivity parameters, including background input and E->E, E->I, I->E, I->I weight gains, within parameter value ranges constrained biologically. Each simulated second required approximately 1 hour on 96 supercomputer cores. For the evolutionary optimization we ran 100 simultaneous simulations (9,600 cores) every generation. To the best of our knowledge, this is the first time evolutionary optimization has been successfully used for large-scale biophysically-detailed network models.
We will use our model to determine mechanistic origins of spatiotemporal neuronal oscillatory patterns observed in vivo using an iterative modeling data-analysis process . At the end of the process, to confirm model predictions, we will use targeted deep brain electrical microstimulation and pharmacological manipulations.
Funding: NIH NIDCD R01DC012947, U24EB028998, U01EB017695, DOH01-C32250GG-3450000, Army Research Office W911NF-19-1-0402
Video Stream
