Google Meet session: meet.google.com/bko-hsja-qee
Fleur Zeldenrust,
Chao Huang,
Prescilla Uijtewaal,
Bernhard Englitz,
Tansu CelikelThe responses of excitatory pyramidal cells and inhibitory interneurons in cortical networks are shaped by each neuron's place in the network (connectivity of the network) and its biophysical properties (ion channel expression [1]), which are modulated by top-down neuromodulatory input, including dopamine. Using a recently developed ex-vivo method [2], we showed [3] that the activation of the D1 receptor (D1R) increases the information transfer of fast spiking, but not regular spiking, cells, by decreasing their threshold. Moreover, we showed that these differences in neural responses are accompanied by faster decision-making on a behavioural level. However, how the single-cell changes in spike responses result in these behavioural changes is still unclear. Here, we aim to bridge the gap between behavioural and single cell effects by considering the effects of D1R activation on a network level.
We took a 3-step approach and simulated the effects of dopamine by lowering the thresholds of inhibitory but not excitatory neurons:
1. Network construction. We created a balanced network of L2/3 and L4 of the barrel cortex, consisting of locally connected integrate-and-fire neurons. We reconstructed the somatosensory cortex in soma resolution ([4], Fig 1A), and adapted the number and ratio of excitatory and inhibitory neurons and the number of thalamic inputs accordingly. 2. Activity of the balanced state. The adaptations in the neural populations and connectivity resulted in a heterogeneous asynchronous regime [5] in L2/3, with highly variable single-neuron firing rates and suggesting a functional role of stimulus separation, and a ‘classical’ asynchronous regime in L 4, with more constant firing rates and suggestive of an information transmission role (Fig 1B). 3. Functional effects. We used a spike-based FORCE learning [6,7] application, trained on either a gap-crossing task (data from [8]) or on a pole detection task (publicly available data from [9], Fig1C). We compared the results against a benchmark test consisting of a 3-layer deep neural net with a recurrent layer.
References[1] Azarfar A, Calcini N, Huang C, et al. Neural coding: A single neuron’s perspective. Neurosci Biobehav Rev 2018;94:238–47.
[2] Zeldenrust F, de Knecht S, Wadman WJ, et al. Estimating the Information Extracted by a Single Spiking Neuron from a Continuous Input Time Series. Front Comput Neurosci 2017;11:49.
[3] Calcini N, Bijlsma A, Zhang Y, et al. Cell-type specific modulation of information transfer by dopamine. Cosyne Abstr. 2019 Lisbon PT, 2019.
[4] Huang C, Zeldenrust F, Celikel T. DepartmentofNeurophysiology/Cortical- representation-of-touch-in-silico. GitHub 2019. https://github.com/DepartmentofNeurophysiology/Cortical-representation-of- touch-in-silico (accessed March 2, 2020).
[5] Ostojic S. Two types of asynchronous activity in networks of excitatory and inhibitory spiking neurons. Nat Neurosci 2014;17:594–600.
[6] Sussillo D, Abbott LF. Generating Coherent Patterns of Activity from Chaotic Neural Networks. Neuron 2009;63:544–57.
[7] Nicola W, Clopath C. Supervised learning in spiking neural networks with FORCE training. Nat Commun 2017;8:1–15.
[8] Azarfar A, Zhang Y, Alishbayli A, et al. An open-source high-speed infrared videography database to study the principles of active sensing in freely navigating rodents. GigaScience 2018;7.
[9] Peron SP, Freeman J, Iyer V, et al. A Cellular Resolution Map of Barrel Cortex Activity during Tactile Behavior. Neuron 2015;86:783–99.
