Visual physiology of the Layer 4 cortical circuitin silico

Abstract

19Despite advances in experimental techniques and accumulation of large datasets concerning the 20 composition and properties of the cortex, quantitative modeling of cortical circuits under in-vivo-like 21 conditions remains challenging. Here we report and publicly release a biophysically detailed circuit 22 109 future, more sophisticated studies of all cortical layers. To enable this, we make the software code, the 110 model, and simulation results publicly available (see SI). 111 112 113 RESULTS 114 Construction and optimization of the model 115The network ( Fig. 1a, b) consisted (see Methods) of models of individual neurons [30] from an early 116 version of the Allen Cell Types Database [31], employing compartmental representation of somato-117 dendritic morphologies (~100-200 compartments per cell) and 10 active conductances at the soma that 118 enabled spiking and spike adaptation. Although recent additions to the Allen Cell Types Database 119include models of neurons with active conductances in the dendrites as well, those models are very 120 computationally expensive, which was prohibitive for the breadth of our study (see below). In addition, 121in terms of somatic spike output, the current versions of such models do not exhibit much better 122 performance than the models with active conductances restricted to the soma [31], and, thus, we used 123 these latter, cheaper models. The cells were distributed uniformly in a cylinder ~400 m in radius, 124representing the central portion of V1, and 100 m height with density of 200,000 mm -3 [32]. Five single 125 neuron models represent five "types" of neurons -three major excitatory groups as determined by Cre-126 lines (Scnn1a, Rorb, Nr5a1, 85% of all cells) and two groups of parvalbumin-positive fast-spiking 127 5 interneurons (15%, denoted as PV1 and PV2), which form the majority of interneurons in L4 [33]. All 128 10,000 biophysical cells were exact copies of these five models. The cell models correspond to regular-129 spiking excitatory cells and fast-spiking interneurons (PV+); whereas non-PV+ interneurons do exist in 130 L4, they are a relative minority [33], and therefore were neglected for simplicity. Furthermore, 35,000 131 much simpler leaky-integrate-and-fire (LIF) neurons, with only two groups -excitatory and inhibitory -132were placed around the biophysically detailed "core" to prevent boundary artefacts (see SI). The 133 complete model accounted for over half of V1 L4 cells (45,000 out of ~70,000). Below, we primarily 134 focus on the properties of the biophysical core circuit. 135Three independent instantiations were generated using different random seeds. The connectivity and 136inputs into these three model instantiations were distinct, but all followed the rules described below. 137All simulations were performed using the python 2.7 code (an early version of the BioNet package [35]) 138 employing NEURON 7.4 [34]. 139

Systematic Integration of Structural and Functional Data into Multi-Scale Models of Mouse Primary Visual Cortex

“…Recent experiments indicate that, besides connection probability, the amplitude (strength) of E-to-E synaptic connections in L2/3 also exhibit a like-to-like dependence (Cossell et al, 2015;Lee et al, 2016). In earlier work, we found these to be even more important for response tuning than connection probability rules (Schaub et al, 2015;Arkhipov et al, 2018). A similar like-to-like rule for synaptic strength (but not connection probability) has been reported for I-to-E connections (Znamenskiy et al, 2018).…”

“…The LGN module is composed of spatio-temporally separable filter units (released publicly via the Brain Modeling ToolKit, github.com/AllenInstitute/bmtk) fitted to electrophysiology recordings from mouse LGN (Durand et al, 2016). In a substantial elaboration over our previous work (Arkhipov et al, 2018), we developed filters for four classes of experimentally observed functional responses (Piscopo et al, 2013;Durand et al, 2016): sustained ON, sustained OFF, transient OFF, and ON/OFF (the latter is related to the DS/OS class of (Piscopo et al, 2013)). These four filter groups are further subdivided according to their maximal response to drifting gratings of different temporal frequencies (TF) ( Fig.…”

“…The weights of this background were adjusted with the recurrent connectivity in place, to ensure that the baseline firing rates of all neurons match experiments (see below). While more sophisticated models of background can be implemented (e.g., Arkhipov et al, 2018), this simple one sufficed for our purpose.…”

“…Cortical circuits feature extensive recurrent connections which amplify thalamocortical inputs and shape cortical computations (Douglas, Martin and Whitteridge, 1989;Douglas et al, 1995;Douglas and Martin, 2007;Lien and Scanziani, 2013;Arkhipov et al, 2018). Despite many studies (e.g., (Cauli et al, 1997;Dantzker and Callaway, 2000;Beierlein and Connors, 2002;Thomson et al, 2002;Beierlein, Gibson and Connors, 2003;Mercer et al, 2005;Song et al, 2005;West et al, 2005;Yoshimura, Dantzker and Callaway, 2005;Lefort et al, 2009;Hofer et al, 2011;Ko et al, 2011;Levy and Reyes, 2012;Olsen et al, 2012;Pfeffer et al, 2013;Vélez-Fort et al, 2014;Bortone, Olsen and Scanziani, 2014;Cossell et al, 2015;Jiang et al, 2015)), data on the exact patterns and magnitude of V1 recurrent connectivity remains sparse, and no resource exists that comprehensively characterizes all connections under standardized conditions.…”

“…Simulating cortical circuits has a long history (e.g. (Wehmeier et al, 1989;Zemel and Sejnowski, 1998;Troyer et al, 1998;Krukowski and Miller, 2001;Traub et al, 2005;Zhu, Shelley and Shapley, 2009;Potjans and Diesmann, 2014;Markram et al, 2015;Arkhipov et al, 2018;Joglekar et al, 2018;Schmidt et al, 2018;Antolík et al, 2019;Schwalger and Chizhov, 2019)), with models incrementally building upon their predecessors. The simulations described here are a further instance of this evolution toward digital simulacra that predict new experiments, are insightful, and ever more faithful to the vast complexity of cortical tissue, in particular its heterogeneous neuronal cell classes, connections, and in vivo activity.…”