Emergence of spontaneous assembly activity in developing neural networks without afferent input

Abstract

Spontaneous activity is a fundamental characteristic of the developing nervous system. Intriguingly, it often takes the form of multiple structured assemblies of neurons. Such assemblies can form even in the absence of afferent input, for instance in the zebrafish optic tectum after bilateral enucleation early in life. While the development of neural assemblies based on structured afferent input has been theoretically well-studied, it is less clear how they could arise in systems without afferent input. Here we show that a recurrent network of binary threshold neurons with initially random weights can form neural assemblies based on a simple Hebbian learning rule. Over development the network becomes increasingly modular while being driven by initially unstructured spontaneous activity, leading to the emergence of neural assemblies. Surprisingly, the set of neurons making up each assembly then continues to evolve, despite the number of assemblies remaining roughly constant. In the mature network assembly activity builds over several timesteps before the activation of the full assembly, as recently observed in calcium-imaging experiments. Our results show that Hebbian learning is sufficient to explain the emergence of highly structured patterns of neural activity in the absence of structured input.

Autonomous emergence of connectivity assemblies via spike triplet interactions

“…To understand the emergence of such non-random connectivity, a growing body of theoretical and computa-24 tional work has been developed to link connectivity architecture to the coordinated spiking activity of neurons, 25 especially in recurrent networks [28][29][30][31][32][33][34][35][36][37][38][39][40][41]. These studies can be divided into two classes: those that examine the 26 influence of externally structured input on activity-dependent refinement [42][43][44][45], and those that investigate the 27 autonomous emergence of non-random connectivity in the absence of patterned external input, purely driven by 28 emergent network interactions [5,6,46]. Specifically, assemblies in recurrent networks can be imprinted based 29 on internally-generated network interactions [6] or through rate-based plasticity where inputs with higher firing 30 rates to subsets of neurons strengthen recurrent connections [47,48]; assemblies can also be initially determined 31 by externally patterned input but maintained by internal correlations [49].…”

“…Finally, modularity is a graph-theoretic measure that describes how strongly a network can be divided into 317 modules, by comparing the relative strengths of connections within and outside modules to the case when 318 the network had weights chosen randomly [94,97,98]. Recently, it was shown that even in models with rate-319 based dynamics, increasing modularity amplifies the recurrent excitation within assemblies evoking spontaneous 320 activation [46]. Thus, it becomes a relevant quantity for characterizing the functional organization of the 321 network.…”

Graph analysis of looming-selective networks in the tectum, and its replication in a simple computational model

“…This match between the model and the experiment suggests that at the very least, our model captured the nature of network development under the influence of synaptic competition and spike-time-dependent plasticity (STDP). Synaptic competition promoted connectivity in weakly connected neurons, while "punishing" overconnected cells, which created light-frame, openwork graph structures (Fiete et al, 2010), and STDP coordinated activity within sub-networks, increasing modularity (Stam et al, 2010;Litwin-Kumar and Doiron, 2014), similar to how it was previously described for other types of plasticity (Damicelli et al, 2018;Triplett et al, 2018). We did not observe changes in the number of neuronal ensembles (Avitan et al, 2017;Pietri et al, 2017), but we believe this is because our experiments were not suited for ensemble detection, as we worked with strong shared inputs that reliably activated almost every neuron in the network.…”

“…For example, we wondered whether it was important to assume that plasticity was stronger during actual collisions, or whether looming selectivity would develop if instead of looming stimuli we used more general visual stimuli. We also wondered whether structured sensory flow is necessary for the emergence of selectivity (Triplett et al, 2018).…”

“…Conversely, they will quickly habituate to actual stimuli, but at the same time, they will have no trouble maintaining synaptic connections required for stimulus detection (Litwin-Kumar and Doiron, 2014). If however intrinsic plasticity is too slow, detection networks will find it easier maintaining "optimal" levels of sensitivity, but may have trouble keeping synaptic connections between stimulus presentations intact (Triplett et al, 2018), as in the absence of spontaneous replay these synaptic connections may erode. A potential solution to this problem is to cycle the network through distinct phases, with different contributions of synaptic and intrinsic plasticity, similar to how we did it in the model.…”