The meso-connectomes of mouse, marmoset, and macaque: network organization and the emergence of higher cognition

Abstract

The recent publications of the inter-areal connectomes for mouse, marmoset, and macaque cortex have allowed deeper comparisons across rodent vs. primate cortical organization. In general, these show that the mouse has very widespread, "all-to-all" inter-areal connectivity (i.e. a "highly dense" connectome in a graph theoretical framework), while primates have a more modular organization. In this review, we highlight the relevance of these differences to function, including the example of primary visual cortex (V1) which, in the mouse, is interconnected with all other areas, therefore including other primary sensory and frontal areas. We argue that this dense inter-areal connectivity benefits multimodal associations, at the cost of reduced functional segregation. Conversely, primates have expanded cortices with a modular connectivity structure, where V1 is almost exclusively interconnected with other visual cortices, themselves organized in relatively segregated streams, and hierarchically higher cortical areas such as prefrontal cortex provide top-down regulation for specifying precise information for working memory storage and manipulation. Increased complexity in cytoarchitecture, connectivity, dendritic spine density, and receptor expression additionally reveal a sharper hierarchical organization in primate cortex. Together, we argue that these primate specializations permit separable deconstruction and selective reconstruction of representations, which is essential to higher cognition.

Keywords: cortex; hippocampus; primate; rodent; working memory.

Fig. 1

Normalized λ as a function of the cubic root of gray matter volume. The EDR exponential decay parameter λ, usually expressed in mm⁻¹, is here multiplied by the maximal inter-areal distance (mm) of the atlas, rendering it dimensionless, and therefore comparable across species. This dimensionless λ increases with brain size, showing that the exponential decay of the EDR becomes sharper as brain size increases.

Fig. 2

Primary sensory and motor areas become more disconnected from other primary cortical areas as brain size increases across species. (A) Lateral view drawings of mouse, marmoset and macaque brains, shown at the same scale. Colors indicate analogous primary areas across the three species for which connectivity data are available: green, visual; orange, somatosensory; blue, motor; red, auditory; and pink, ventromedial prefrontal cortex (not shown). Bar scale, 1 cm. (B) Reduced retrograde tract-tracing, inter-areal connectivity matrices for each species, resulting from injections in primary visual, somatosensory, and motor (columns), and showing 5 source areas (rows): V1, primary visual; AUDp/AUCore/Core, primary auditory; SSP/S1, primary somatosensory; mop/M1/F1, primary motor, PL/IL prelimbic/infralimbic (mouse) and its primate equivalent area 25/32. Each colored entry represents a connection from a row area to a column one. Connection weights are expressed in log10 scale, ranging 6 orders of magnitude, from dark brown (very weak) to bright yellow (very strong). Gray represents absent connections and hatched squares, self-connections. (C) Circular graph representation of connectivity data shown in (B), with color-coding from (A). Injected areas are depicted with needles. Arrow color intensity and width indicate weights: thin pale arrows: weak connections; thick dark blue arrows: strong connections.

Fig. 3

Effect of brain size on the topology of the inter-areal network. In smaller brains with lower wiring cost, an EDR-based brain with a smaller (normalized) λ parameter leads to an “all-to-all,” (near-) complete graph (i.e. very high density), as shown by retrograde tract tracing data. In such brains, information propagates from sensory primary areas to the entire network very quickly (blue connections), with little room available for hierarchical information processing or segregation of function. In bigger brains such as great apes, cetaceans, or elephants, the increased wiring cost leads to a larger (normalized) λ and a correspondingly sharper EDR. Density falls drastically (here chosen at 0.07 for illustration), leaving place to a much sparser network. We posit that, in such sparse network, we should expect a k-NN topology akin to a lattice, where each area is primarily connected to its direct neighbors and only a few longer-range connections otherwise remaining. There, information would spread gradually, from neighbor to neighbor, allowing both hierarchical processing and segregation of function (blue and orange connections). Intermediate size primate brains, such as marmosets or macaques, would fall in-between those two ends of the spectrum.

Fig. 4

Examples of multimodal convergence and functional segregation in the macaque cortex. (A) Wiring diagram depicting inputs to perirhinal (areas 35 and 36) and parahippocampal cortices (areas TH and TF), which are major inputs to the entorhinal cortex (ERC), the main cortical entry point into the hippocampus (HPC). Inputs to the perirhinal and parahippocampal cortices are higher order association areas, and often polymodal. The perirhinal areas are more associated with feature-specific inputs and functions, while the parahippocampal areas are more associated with spatially related inputs and functions. Adapted from Suzuki and Amaral (2004). (B) The separation of sensory streams for spatial vs. feature information for both visual and auditory information continues into the PFC, terminating in dorsal (spatial) vs. ventral (feature) zones of the macaque dlPFC as conceived by Patricia Goldman-Rakic; adapted from a drawing by Mark S. Williams in (Arnsten 2003). Recent work suggests that auditory zones extend to the frontal pole and medial surface as well (Medalla and Barbas 2014).

Fig. 5

Frontopolar- “top-down” frontopolar projections to auditory association areas of the superior temporal gyrus. Area 10 axons terminate in a feedback pattern, with smaller frequent boutons in layer I and larger, sparser boutons in mid-to-deep layers. Terminations in layer I likely interact with disinhibitory inhibitory neurons (CR, calretinin), while in layer II to III, they interact more frequently with inhibitory neurons that target the apical dendrites (CB, calbindin) or pyramidal neurons; in mid-deep layers, they are more likely to target perisomatic inhibitory neurons (PV, parvalbumin). Adapted from Figs 4 and 5 of Medalla and Barbas (2014).

Fig. 6

Molecular gradients aligned with the cortical hierarchy in primates. (A) As described in the text, research has shown increasing expression of GRIN2B, CALB1, DRD1, and HTR1A encoding the GluN2B subunit of the NMDAR that closes slowly and fluxes high levels of calcium, the calcium-binding protein, calbindin, the dopamine D1 receptor, and the serotonin 5HT_1A receptor, respectively. GRIN2B (NMDAR-GluN2B) are essential to persistent dlPFC neuronal firing during working memory. (B) The cortical hierarchy across the dorsal stream with increasing timescales across regions, from V1 to MT, to LIP to dlPFC to ACC (anterior cingulate), areas often used to compare molecular expression levels. (C) The levels of GRIN2B expression in dlPFC increase across primate brain evolution, adapted from Muntané et al. (2015). Figure based on Murray et al. (2014); Burt et al. (2018); Arnsten et al. (2021); Froudist-Walsh et al. (2021); Froudist-Walsh et al. (2023).

Fig. 7

A working model of how dopamine actions at D1R in layer III of dlPFC may sculpt network inputs and refine the contents of working memory. Dopamine D1R are concentrated on dendritic spines in layer III dlPFC, often at extrasynaptic locations where they are colocalized with ion channels that weaken network connectivity, such as HCN-slack channels (Paspalas et al. 2013; Gamo et al. 2015; Wu et al. 2024). Thus, weaker (nonpreferred) inputs may be differentially gated out. Enhancement of lateral inhibition from PV interneurons may also contribute to the refinement of representations in working memory. Based on Vijayraghavan et al. (2007) and Arnsten et al. (2021).