Table 2.

Table of the heuristics and regularities used to construct the model along with starting points for extensions, if applicable

Feature	Heuristic	Argument	Starting points for extensions
Population sizes	Neuron densities of areas with missing data equal the mean neuron density for areas of the same architectural type.	Neuron density varies systematically with architectural type
Population sizes	Areas MIP and MDP have architectural type 5	Their neighboring area PO, similarly involved in visual reaching (Johnson et al. 1996; Galletti et al. 2003), is of type 5 (Hilgetag et al. 2016)
Population sizes	Total thickness and relative laminar thicknesses for areas with missing data are linearly predicted from the logarithm of their overall neuron density	This follows observed gradients. The increase in relative L4 thickness with log neuron density is consistent with L4 thickness entering into the definition of the architectural types
Population sizes	The fraction of excitatory neurons in each layer is identical across areas	This provides a simple rule across areas, for lack of systematic area-specific data	Beaulieu et al. (1992) report similar values for layer-specific fractions of inhibitory neurons in macaque V1. Gabbott and Bacon (1996) report layer-specific fractions of inhibitory neurons in macaque medial prefrontal cortex differing from the values of Beaulieu et al. (1992)
Local connectivity	We assume an underlying Gaussian model for the local connection probability	This ansatz provides consistency with the derivations of Potjans and Diesmann (2014)	Markov et al. (2011) report an exponential decay of locally labeled neurons with distance from the injection site. With assumptions on cell density, this enables deriving a non-Gaussian distance-dependent connection probability
Local connectivity	Population pairs have the same relative indegrees as in the model of Potjans and Diesmann (2014)	This follows the notion of a canonical microcircuit (Douglas et al. 1989; Douglas and Martin 2004), for lack of comprehensive species- and area-specific data	Beul and Hilgetag (2015) suggest a canonical microcircuit for agranular cortical areas, which in our model includes area TH
Local connectivity	The relative amount of local synapses is constant across areas	The fraction of labeled neurons intrinsic to the injected area found by retrograde tracing is approximately constant
Long-range connectivity	All cortico-cortical connections originate and terminate in the $1{\mathrm{mm}}^2$ patches covered by our model	Since we do not explicitly include spatial dependence of connections, we opt for a simple model for cortico-cortical connections	Cortico-cortical connections exhibit divergence and convergence (Colby et al. 1988; Salin et al. 1989; Gattass et al. 1997; Markov et al. 2014b)
Long-range connectivity	All cortico-cortical connections are excitatory	This simplification approximates the finding that the large majority of cortico-cortical projections are excitatory	A small fraction of cortico-cortical connections in monkey (Tomioka and Rockland 2007) and other species (McDonald and Burkhalter 1993; Gonchar et al. 1995; Fabri and Manzoni 1996, 2004; Tomioka et al. 2005; Pinto et al. 2006; Higo et al. 2007) are inhibitory
Long-range connectivity	Neurons in all source areas form the same number of synapses in each target area	This assumption allows us to directly translate FLN into synapse numbers	There is evidence that numbers of cortico-cortical synapses per neuron differ between feedback and feedforward connections (Rockland 2003)
Long-range connectivity	The probability for a postsynaptic neuron to form a cortico-cortical synapse in a specific layer is constant across areas.	For lack of data in areas besides V1, we take the computed values from the Binzegger et al. (2004) data as representative across the model
Long-range connectivity	The probability for a synapse to be established on a neuron of a given type is proportional to the length of the dendrites of the neuron type in the given layer	This heuristic is a version of Peters’s rule, which has been shown to have reasonably wide validity at the population level (Rees et al. 2016)
Long-range connectivity	The relative number of synapses sent by supragranular neurons is filled in based on the logarithmic ratio of overall cell densities in the two participating areas	This follows the observed relation between SLN and the log ratio of overall cell densities in combination with interpreting ratios of labeled neurons as ratios of formed synapses
Long-range connectivity	The level of SLN predicts the type of laminar termination pattern	This follows the observed relation between SLN and termination pattern
Long-range connectivity	Feedforward and feedback pathways are not separate within layers: individual neurons can send both types of connections	This heuristic is used to avoid the added complexity that would result from further subdivisions of the neural populations	A finer definition of laminar pathways may be achieved via a dual counterstream organization (Markov et al. 2014b)