NETFORM - Neuronal Network Formation through Reciprocal Interactions between Activity and Structure


Return

Project B. Network connectivity and activity
Creating a model of network development to study the interdependent development of network connectivity and activity under the influence of local activity-dependent processes.

Aim.
Creating a model of network development to study the interdependent development of networkconnectivity and activity under the influence of local activity-dependent processes. Relevance. Due to activity-dependent process that influences synaptic connectivity (neurite outgrowth, synaptic strengths, intrinsic neuronal excitability), a reciprocal interaction exists between neuronal activity and synaptic connectivity. Activity thereby appears to modulate synaptic connectivity so as to maintain the average electrical activity of a neuron at a particular level (homeostatic regulation). Decreased neuronal activity causes an increase in the strength of all excitatory synapses onto the neurons, whereas increased activity reduces the strength [41,42]. We hypothesize that the reciprocal interaction between neuronal activity (fast dynamics) and synaptic connectivity (slow dynamics) organizes networks and renders them robust to ongoing plasticity. An important open question [25] that we address is whether homeostatic, activity-dependent mechanisms at the local synapse or neuron level are capable of achieving homeostasis of global network activity. Local tuning rules may be able to keep the activity of a single neuron at a desired level, but are they sufficient to maintain adequate network performance [48,19,20]? Our results will be instrumental in interpreting data sets on synaptic connectivity and network activity (see under Validation).

Background.
To study the interdependent development of network connectivity and activity, we need a model that incorporates neuronal activity; synaptic connectivity; and activity-dependent mechanisms of neurite outgrowth, synaptic strengths and intrinsic neuronal excitability. Existing models do not suffice because they either model axons and dendrites as growing circular neuritic fields, so that synaptic connectivity is not well represented [48,51,2]; implement only activity-dependent neurite outgrowth but no changes in synaptic strengths or intrinsic neuronal excitability [48,51,8,2]; implement only activity-dependent neuronal excitability [14]; or do not use spiking models of neuronal activity [48,51]. We are confident that our new model will give important 6 insights, because even these more simplified models [47,48,49,51] yield a rich repertoire of emergent properties, including morphological differentiation between excitatory and inhibitory cells, self-repair, and possibly critical connectivity [2].

Approach.
We will construct a model of network formation that contains activity-dependent, homeostatic mechanisms at all three spatial levels: neurite outgrowth (structural connectivity), synaptic strength, and intrinsic neuronal excitability. For structural connectivity, we will describe neuronal morphology significantly more detailed than in the neuritic field approach [48,51], but much more simple than in project A, because it should allow us to study the interdependent development of connectivity and activity patterns in large networks. A level of description similar to that in [8] will be considered, in which each neuron has separate axonal and dendritic elements that can make synaptic connections to other cells, without imposing unrealistic spatial constraints as in [48,51]. Changes in synaptic strengths will be implemented as spike-timing dependent plasticity [34], with overall up- and down-scaling of synaptic strength to represent homeostatic synaptic plasticity [41,42]. Activitydependent regulation of intrinsic neuronal excitability could be represented by an adaptable firing threshold. For neuronal activity, we will use a spiking model, such as a Hodgkin-Huxley or integrate-and-fire model. Cells will have a low background firing rate, so that activity generated by the network itself will drive development. Insights from projects A and C, on the relation between neuronal morphology and synaptic connectivity, will help define the more abstract spatial rules that are necessary here to model connectivity formation. The statistical methods from subproject C, on deriving underlying network connectivity from firing activities, will be used to validate the development of connectivity in the model with that in developing cultures of cortical neurons (data sets Van Pelt; [53,54]).

Research Questions and Validation.
  1. Can activity-dependent processes operating at the local level of synapse or neuron achieve homeostasis of activity at the network level? To test this, we will do perturbation studies of network connectivity and activity.
  2. Can these activity-dependent processes form networks with critical (and possibly small-world) connectivity?
  3. Does electrical activity hypothesized to be associated with critical connectivity, such as neuronal avalanches [5,53] and activity with long-range temporal correlations [24], arise?
  4. For all above, is the combination of activity-dependent processes essential, or can individual process already produce the desired results?
We will validate the development of activity and connectivity (as obtained by the reverse engineering approach in project C) with that in the unique data we have on developing cultures of cortical neurons (Van Pelt; [53,54]). To assess whether our model can generate electrical activity hypothesized to be associated with critical connectivity, we will consider data sets on neuronal avalanches [5] and activity patterns with long-range temporal correlations in cortex and cortical brain slices ([5]; and data Linkenkaer-Hansen [24]). Additionally, we may provide insight into how activity-dependent loss of small-world connectivity, as observed in epilepsy [38,28], could arise.