The space-time fabric of brain networks | From space to space-time in the brain
October 28, 2019: Scientists at the Bernstein Center Freiburg (BCF) and KTH Stockholm found simple rules that enable networks in the brain to generate sequential spiking activity, forming the basis of behavior. Their findings have now been published in PLoS Computational Biology.
Figure Legend see below. Click to view full-size image.
Every behavior involves a specific well-ordered sequence of actions. How the brain is capable of generating such sequential actions is one of the fundamental questions in neuroscience. One of the basic requirements for meaningful behavior is that networks in the brain generate well-defined sequences of neuronal activity. In the past few years, experiments have revealed that animal behavior is accompanied by sequential activity of neurons in various brain regions. Following-up such observations, theoreticians have proposed several mechanisms that could explain the emergence of sequential activity of neurons. These mechanisms primarily rely on supervised learning, in which network connectivity is adjusted according to some learning rule or synaptic plasticity rule to generate the desired sequential activity.
“It is obvious that neuronal networks can be ‘trained’ to generate sequential activity. But we know that not all behavior is learned. Innate behavior suggests that the brain generates sequential activity without learning and training” said Arvind Kumar (KTH Stockholm), the lead author of the study. But how can an untrained brain generate well-ordered activity sequences?
To address this question, he teamed up with Sebastian Spreizer, a PhD student at the BCF and Prof. Ad Aertsen, now an Emeritus Professor at the BCF and the University of Freiburg. Together they found two conditions that enable a neuronal network to generate sequential activity: (1) neurons project a small fraction of their outputs to a preferential direction and (2) the preferred directions of neighboring neurons are similar. “The first condition implies that the connectivity is anisotropic and the second condition implies that connectivity is spatially correlated. That is, ‘correlated spatial anisotropic connectivity’ is the key to generating sequential activity in neuronal networks” explained Sebastian Spreizer, the first author of the study. When the network is wired according to these rules, the network connectivity creates an activity landscape akin to geographical landscapes with hills and valleys. Following this metaphor, sequences of neuronal activity are similar to the water streams and rivers in such landscape. Thus, in this network model, small variations in the space of neuronal connectivity create both spatial and temporal sequences of neuronal activity.
We know that the morphology of neurons in the brain is highly anisotropic. Moreover, there are several reasons to assume that neighboring neurons are anisotropic in a similar fashion. However, to verify the model, the shapes and connectivity of neighboring neurons need to be measured. “But, neuron morphology is not the only way to generate such networks. Interestingly, neuromodulators such as dopamine can create correlated spatial anisotropic connectivity in the brain in a dynamic, context-dependent manner.” said Ad Aertsen.
This new mechanism gives the network an ability to create various sequences of neuronal activity. However, not all of these will be functionally relevant and rewarding to the animal. Hence, the authors of this study hypothesize that, once the network become a sequence generating machine, learning mechanisms can kick in to select the rewarding sequences and unlearn the others. Thus, correlated spatial anisotropic connectivity should make learning faster and more efficient.
Original Publication
Spreizer S, Aertsen A, KumarA (2019) From space to time: Spatial inhomogeneities lead to the emergence of spatiotemporal sequences in spiking neuronal networks.
PLoS Comput Biol 15(10): e1007432.
Figure Legend
An anisotropic but correlated connectivity pattern of neuron groups in the brain suffices to generate spatio-temporal activity sequences in an otherwise random neuronal network model. Figure: Sebastian Spreizer.
Contact
University of Freiburg
Bernstein Center Freiburg
Hansastr. 9a
79104 Freiburg
Germany
Sebastian Spreizer
E-mail: sebastian.spreizer@bcf.uni-freiburg.de
Prof. em. Dr. Ad Aertsen
E-mail: ad.aertsen@biologie.uni-freiburg.de
KTH Royal Institute of Technology
Prof. Dr. Arvind Kumar
Dept. of Computational Science and Technology
Lindstedtsvagen 5
Stockholm, Sweden
Tel: +46 (8) 790 62 24
E-mail: arvkumar@kth.se