Normalized cortical depth (NCD) as a primary coordinate system for cell connectivity in cortex: experiment and model

Traditionally, cortical microcircuits have been described using layers. Here we argue that there are several advantages to using normalized cortical depth (NCD) instead of layers as a primary reference system. Unlike layer definitions which are variously interpreted, NCD provides a well-defined, consistent and continuous reference system, depending only on two readily-identifiable landmarks: pia (NCD=0) and white matter (NCD=1). It is therefore desirable as a means to identify cells and connections for sharing across experimental datasets. By contrast, in most cases, cortical layers can only be delineated in the tissue after recording is completed. In general, layer identity will differ depending on staining and is subject to alterations due to shrinkage and inadequate marking, and to differences in interpretation across laboratories and over time. For example, L4 has recently been identified in primary motor cortex (M1), which was classically considered to lack this layer (Hooks, 2013 J Neurosci 33:2). NCD is also better suited to represent connectivity, which varies systematically within and across layers. The strength of input from from L2/3 to L5B corticospinal cells in M1 depends on the depth of cell soma (Anderson, 2010 Nat Neu 13:6), and can be approximated as a gaussian function of NCD, with mean=0.51 and std=0.06 (R^2=0.97). Similarly, the normalized strength of input from S2 to L5B CSP cells in M1 can be approximated as a linear function of NCD with slope=-5.6 (R^2=0.7) (Suter, 2015 J Neurosci 35:7). We used data from M1 to develop a multiscale computer model of M1 based on NCD coordinates in NEURON. Use of NCD facilitated employing the experimental data to constrain the model, and enabled wiring according to the depth-dependent connectivity profiles described above. Connectivity was also dependent on cell type. We evaluated dynamical interactions across scales, particularly those involving the the two main types of deep projecting cells (corticospinal and corticostriatal). This was developed as a hybrid model with these 2 major cell types characterized using multicompartment models, optimized to reproduce empirical intrinsic physiology of soma and dendrites, while other network cells were implemented using extended integrate and fire models (eg. Izhikevich models). Use of a hybrid model enabled the simulation of larger scale cortical networks while still exploring the dynamics of complex cells. Incorporating cortical depth dependent features into the model brings us a step closer to capturing the complex multiscale spatiotemporal interactions occurring in cortex, many of which have been largely omitted in the classical canonical microcircuit.