Functional constraints in the evolution of brain circuits

Abstract

Regardless of major anatomical and neurodevelopmental differences, the vertebrate isocortex shows a remarkably well-conserved organization. In the isocortex, reciprocal connections between excitatory and inhibitory neurons are distributed across multiple layers, encompassing modular, dynamical and recurrent functional networks during information processing. These dynamical brain networks are often organized in neuronal assemblies interacting through rhythmic phase relationships. Accordingly, these oscillatory interactions are observed across multiple brain scale levels, and they are associated with several sensory, motor, and cognitive processes. Most notably, oscillatory interactions are also found in the complete spectrum of vertebrates. Yet, it is unknown why this functional organization is so well conserved in evolution. In this perspective, we propose some ideas about how functional requirements of the isocortex can account for the evolutionary stability observed in microcircuits across vertebrates. We argue that isocortex architectures represent canonical microcircuits resulting from: (i) the early selection of neuronal architectures based on the oscillatory excitatory-inhibitory balance, which lead to the implementation of compartmentalized oscillations and (ii) the subsequent emergence of inferential coding strategies (predictive coding), which are able to expand computational capacities. We also argue that these functional constraints may be the result of several advantages that oscillatory activity contributes to brain network processes, such as information transmission and code reliability. In this manner, similarities in mesoscale brain circuitry and input-output organization between different vertebrate groups may reflect evolutionary constraints imposed by these functional requirements, which may or may not be traceable to a common ancestor.

Keywords: canonical microcircuits; cortical evolution; cortical neurodevelopment; neuronal oscillations; predictive coding.

Figure 1

(A) Simplified representation of a six- (left) and three- (right) layer microcircuit. Neurons depicted in blue (open synapses) are excitatory cells, whereas neurons in red (close synapses) are inhibitory ones. Black arrows represent the flow of information across different layers. Both panels adapted from Shepherd (2011). (B) Schematic representation of a canonical microcircuit. Arrows represent connectivity within nodes, ordered spatially according to their anatomical localization. Curved arrows illustrate intrinsic (excitatory and inhibitory) connectivity. Adapted from Douglas and Martin (2004).

Figure 2

(A) Laminar recording on visual cortex using a linear microelectrode array (LMA). (B) Stimulus-evoked current source density (CSD) analysis using laminar probes. LMAs typically depict sinks (cation inflow from the extracellular to the intracellular space, in red) and sources (cation outflow from the intracellular to the extracellular space, in blue) of neuronal activity. This configuration pattern is useful to segregate differential layer activity. (C) Noise correlation amplitude is significantly lower in granular layers as compared with supra and infragranular layers. Black and gray bars denote two different monkeys. (B,C) adapted from Hansen et al. (2012). (D) Decrease of noise correlations amplitude in cyto-architectonically defined intermediate layers of the songbird auditory caudal meso/nidopallium cortex [Asteriks indicate significant differences in correlations between regions (p < 0.05, Kruskal-Wallis test with multiple comparisons correction)]. Adapted from Calabrese and Woolley (2015).

Figure 3

(A) MRI-scan based rendering of a monkey brain implanted with an ECoG grid. Lines indicate the boundaries of recorded regions. Dots indicated 252 recording sites. (B) Granger-causal influence spectra of across all regions of interests separated for top-down (black line) and bottom-up (green line) directions. Spectral top-down influences are mostly at beta frequency-band, whereas bottom-up influences occurs at gamma frequency-band. Asterisks indicate significant differences (p < 0.001, randomization test). (C) Hierarchical ranking of recorded areas according to Markov et al. (2014). Green lines indicate bottom-up influences. Black lines denote top-down influences. (A–C) are modified from Bastos et al. (2015b). (D) Representation of Granger causality influence across layers for gamma (left) and alpha (right) frequency bands. Gamma causal influences depart from L4 to infra and supragranular layers. Alpha is mostly observed in infragranular layers and influences granular and supragranular layers. (E) Laminar profile of spike-field coherence (measured through the CSD-MUA coherence) for gamma (left) and alpha (right) frequency band. Middle and supragranular layers show increased gamma field coherence whereas alpha is concentrated in infragranular layers. (D,E) are modified from van Kerkoerle et al. (2014).