e., equation(10) σ(R)=(1−cϕ)g0(R)+cϕg1(R).corresponding to Equation 1. The transfer of correlations cξ→cϕcξ→cϕ from input currents to potentials is, in a realistic setting (i.e., for frequency dependent current-density filters), nontrivial. A rigorous mathematical treatment of this is beyond the scope of this work. Instead, we investigate the current-potential correlation transfer for different neuron types and synapse distributions numerically (see Results; Figure 4). Below follows a summary of the numerical simulations based on reconstructed morphologies. For tables containing model details and PLX 4720 parameter values, see Tables S2–S4.

Multicompartment neuron models with morphologies from digital cell reconstructions (see below) were randomly positioned in a cylindrical volume with radius 1,000 μm. Each population consisted of 10,000 cells with identical cell morphology but each cell was randomly rotated along the z axis. The somata of all cells in a population were placed at the same cortical depth, chosen

as the midpoint of the corresponding cortical layer. Layer boundaries were derived from Stepanyants et al. (2008). The same x mTOR inhibitor and y coordinates were used for populations of the three different cell types to remove variability due to the exact cell positioning when comparing different cell types. See Supplemental Experimental Procedures for details. We used morphological reconstructions of L3 pyramidal, L4 spiny stellate, and L5 pyramidal neurons (Mainen and Sejnowski, 1996) downloaded from ModelDB (http://senselab.med.yale.edu/modeldb)

from which we removed axon compartments and active conductances (making the models passive). For passive parameters and details on spatial segmentation, see Supplemental Experimental Procedures. Simulations were performed with a time resolution of 0.0625 ms and resulting data was stored with a time also resolution of 1.0 ms. Simulations were in all applications run for a time period of 1200 ms were the first 200 ms were removed before analysis to avoid any upstart effects in the simulations. Postsynaptic currents (PSCs) were modeled as α-currents triggered by the arrival of presynaptic input spikes (for details, see Supplemental Experimental Procedures). For the results shown in Figure 2, Figure 3, Figure 4, Figure 5 and Figure 7, only excitatory synapses (EPSCs) were used, while both excitatory and inhibitory synapses (IPSCs) were used in the simulations with laminar-network input (Figure 6; see below). The amplitude of a single IPSC was four times stronger than an EPSC. Note: since the neuron models are linear with respect to the amplitude of current injection, the results will not change with other values of the input current (as long as the relative values for excitatory and inhibitory synapses are fixed) except a rescaling of the resulting LFP amplitudes.