In the brain information is encoded through the spatiotemporal activity patterns of neuronal ensembles and correspondingly, the network of connections is an integral part of the processing algorithm itself. Although mammalian brains show a massive, 5 order of magnitude variation in weight, they present common processing features. This suggests that cortical computing happens across scalable architectures, possibly sharing common architectural principles, critical for efficient and hierarchically modular information processing. This project, with combined experimental and computing/theoretical components, has set out to generate extensive and updated physical connectivity databases between brain functional areas in both cynomolgus monkey (macaque) and mouse and to analyze these networks in search for architectural network invariants. It has also generated a much improved and validated areal atlas for the money. The experimental data is an extensive record of high accuracy retrograde tract-tracing labeling in both species, on a 40x91 interareal matrix (soon to become 50x91) for the macaque, and 19x47 matrix for the mouse. Since the generation of this data is invasive, time consuming and costly, the theoretical component of the project has developed novel link-prediction and network imputation algorithms for tract-tracing type data. These interareal networks are spatially embedded, highly dense, directed, weighted and because of that, most network analysis and inference algorithms (usually developed for binary, undirected networks) do not apply to them. We have demonstrated that there is significant amount of predictability (about 80% on average) in the weighted, directed, spatially embedded network of the mammalian cortex (both monkey and rodent), indicating the existence of structural regularity (as opposed to randomness) in cortical networks – aka architectural network invariants. Using our imputation algorithms, we have generated the full interareal networks for both species (91x91 in the macaque and 47x47 in the mouse), which have been used as part of whole-brain dynamical models to study cognitive functions and states of consciousness, by other groups. Our more extended datasets have further strengthened our earlier observations that the mesoscale structure of the mammalian brain is strongly shaped by the Exponential Distance Rule (EDR). By performing sub-parcellations of the early visual areas and additional injections into them (central and peripheral), we have shown that the EDR is a strong organizer of the feedback connections to these areas. Connection strengths for most projections vary significantly with eccentricity in a systematic fashion with respect to distance and origin; whereas projections to central and upper visual fields are significantly stronger from ventral stream areas, peripheral and lower field projections are stronger from the dorsal stream. As part of our structural analysis efforts, we also developed novel link community detection algorithms for tract-tracing neuronal network datasets, revealing a structured and hierarchical organization within the mammalian cortex, allowing for species comparison and differentiation.

The claustrum is a cortex-like area, referred to it as the “conductor of consciousness” (Crick, Koch, 2005). Our question was whether our macaque data on the connectivity from the claustrum to the cortex (obtained from the 40 injections into the cortex) would indeed be consistent with the cortical organizer hypothesis of the claustrum, which has been confirmed. For the first time, we have been able to also perform retrograde tract-tracing injections into the claustrum. Combined with the label counts in the claustrum due to injections in 40 cortical areas, we have generated a physical network of connections revealing in both directions the connectivity of the claustrum with the cortex. This data will now allow us to better understand the role of the claustrum, a perennial question in neuroscience. For example, the telencephalic nature of the claustrum is now confirmed by the numerous supra-granular layer neurons projecting to the claustrum, particularly in frontal areas.

We have provided a simple and fundamental mechanism for the functional arealization, namely for functional specialization in multitasking networks, which has never been proposed, to the best of our knowledge. We have also introduced a novel network growth mechanism that treats nodal degree as an innate property of the nodes and thus it preserves node degree during network evolution, in-line with F. Jacob’s concept of evolution via ‘tinkering’ (Science 1977). This is a first such model family proposed for network modeling, especially for biological and neuronal networks.

Last Modified: 05/25/2022
Modified by: Zoltan Toroczkai