Alternative view on structural plasticity in neuronal networks: Hebbian associative properties can also result from firing rate homeostasis
Description see below
Neuronal networks in the brain undergo steady remodeling. They continuously create, delete or rewire connections between neurons. This capability is indeed fundamental to all learning and memory processes. Experts call this phenomenon “structural plasticity”. Plasticity processes are generally thought to come in two flavors: Hebbian or homeostatic. Put simply, Hebbian means that growth depends on the correlation between the activity of two neurons – what fires together, wires together. Homeostatic regulation of connectivity, in contrast, is thought to mainly stabilize neuronal signaling and network wiring to avoid runaway activity and unbounded growth. But is it necessarily distinct mechanisms that provide Hebbian and homeostatic functionality? This is the question that PhD student Júlia Gallinaro and Prof. Stefan Rotter from the Bernstein Center Freiburg found a new and interesting answer to.
“It is commonly thought that associative principles in neuronal networks should rely on correlations between the activity of neurons, which also must be somehow represented at the level of individual synapses”, explains Júlia Gallinaro. But what if there is no synapse yet, and a new one has to be formed? “We wanted to find out if associative learning, usually assumed to follow rules first formulated by the psychologist Donald O. Hebb in 1949, could also emerge from a plasticity rule that is based on homeostatic principles alone.” This would imply a totally new perspective on the mechanisms that control network remodeling, both during development and in the adult. “We were surprised to see that it works amazingly well in computer simulations”, Júlia Gallinaro points out.
Homeostasis can implement Hebb’s rule
To systematically address their question, the researchers performed a thought experiment, the outcome of which may inspire neurophysiologists to make new biological experiments. They used methods from computational neuroscience and performed large-scale simulations of spiking neural networks with a structural plasticity rule based on firing rate homeostasis. In a nutshell, they could show that, under these conditions, a subgroup of neurons developed stronger within-group connectivity after having received stronger external stimulation than the other neurons in the network.
Turning to an experimentally already well-documented situation, the scientists then considered the maturation of primary visual cortex of mice in the first weeks after eye-opening. It was previously demonstrated that visual experience leads to the emergence of new and specific connections that reflect the use-dependent fine-tuning of sensory networks.
Júlia Gallinaro and Stefan Rotter could show that such specific connectivity can indeed emerge in a network which is continuously remodeled using a structural plasticity rule that does not explicitly depend on activity correlation. The structural changes triggered by visual input are long-lasting and decay only very slowly when the neurons are exposed again to unspecific external inputs. Obviously, this is a desirable property of any kind of memory.
Make new neuronal friends
Similar to learning rules which explicitly depend on correlations, homeostatic plasticity implements some sort of “social networking” in the brain. “Several neurons signal their interest in creating a new connection, and then those interested are randomly matched together”, says Júlia Gallinaro. The twist is that there is no need to have information about the other neuron’s activity. “As one of our colleagues put it: You can only make friends with people who also want to make friends.“ Neurons therefore tend to form connections with other neurons that are in a similar state at the same time – just as people who move to a new city at the same time and who have similar interests are more likely to make friends.
The networks of our brain reflect our past experiences and prepare us for future challenges. Their detailed structure is established and fine-tuned in an extremely complex but stunningly well-orchestrated process, it is obviously well maintained throughout our entire life, and it may decay with age and during disease. “The rules of growth and plasticity are really essential. Our new findings are just a small piece in this huge puzzle”, concludes Júlia Gallinaro. “Of course, we had to greatly simplify things in our first study in order to understand and demonstrate the basic principle. The complexity of biological brains exposed in experiments, however, will now guide us to refine and improve our description.”
Figure Caption:
Connections between excitatory neurons are subject to structural plasticity, based on firing rate homeostasis. Similar to a thermostat, neurons create and delete incoming and outgoing synapses in order to keep a predefined level of activity. Perturbations lead to network remodeling, which implements associative learning.
Original publication:
Gallinaro JV, Rotter S. Associative properties of structural plasticity based on firing rate homeostasis in recurrent neuronal networks. Scientific Reports 8: 3754, 2018
https://www.nature.com/articles/s41598-018-22077-3
Contact:
Júlia Gallinaro
Bernstein Center Freiburg
University of Freiburg
Hansastr. 9a
79104 Freiburg i.Br.
Tel.: +49 (0)761 203 9313
E-mail: julia.gallinaro@bcf.uni-freiburg.de
Prof. Dr. Stefan Rotter
Bernstein Center Freiburg
University of Freiburg
Hansastr. 9a
79104 Freiburg, Germany
Tel.: +49 (0)761 203 9316
E-mail: stefan.rotter@bio.uni-freiburg.de