Neural networks develop naturally under continuous reciprocal interaction with the environment, which results in early life experiences having an immense impact on adult capacities to function and adapt to the environment (Vygotsky, 1978; Black et al., 1987; Greenough et al., 1987; Sur et al., 1988; Pallas et al., 1990; Jones and Greenough, 1996; Markham and Greenough, 2004; Marom, 2015). It is well established, for example, that sensory inputs during development play an essential role in shaping brain circuitry and neuronal connectivity, as was shown by the classical studies of Hubel and Wiesel in which sensory deprivation enhanced fundamental alterations of primary sensory areas (Hubel and Wiesel, 1977; Hubel et al., 1977; Greenough et al., 1987; Sur et al., 1988). However, the underlying neural principles and mechanisms which impact such developmental interactions remain largely unknown. This state-of-the-art environment-development problem, calls for controlled, multi-level experimental access to neural networks development over a wide range of temporal and spatial scales. The major aim of this work, therefore, is to provide an experimental database for studying how external inputs encountered during development can affect the mature aspects of neuronal network activity—implemented in neural networks developing in-vitro.

Cultures of cortical networks comprise a relevant experimental model for characterizing development of neural activity, without being affected by predefined constraints and intervening processes (Katz, 1993; Wong, 1993; Marom and Shahaf, 2002). The developmental stages and their corresponding time frames are surprisingly similar to those observed in vivo. When a population of cortical neurons is extracted from a rat newborn and allowed to develop for a couple of weeks outside the brain, it forms a network that exhibits complex spontaneous activity. During this process, uncorrelated firing appears at the end of the first week in culture, then evolving to synchronous activations observed as early as 9–12 days in-vitro (Gross, 1979; Habets et al., 1987; Stenger and McKenna, 1994; Kamioka et al., 1996; Corner et al., 2002; Marom and Shahaf, 2002; Morin et al., 2005; Eytan and Marom, 2006; Wagenaar et al., 2006). Synchronous events frequency increase to an average rate of ca. 0.1 Hz, in dependence on a rapid growth in networks connectivity, both in terms of number of connections and connection length, between 5 and 10 days in-vitro (Maeda et al., 1995; Marom and Shahaf, 2002; van Pelt et al., 2005; Wagenaar et al., 2006; Napoli et al., 2014).

By applying electric stimulation to these networks, it is possible to evoke a synchronized response that propagates across extended parts of the network (Maeda et al., 1995; Pinato et al., 1999; Jimbo et al., 2000; Marom and Shahaf, 2002; Wagenaar et al., 2004). Such stimulation has been shown to simulate modulatory effects of sensory inputs, as inducing neural plasticity and changes in network connectivity (Jimbo et al., 1998, 1999; Maeda et al., 1998; Tateno and Jimbo, 1999; Rolston et al., 2007; Bologna et al., 2010; Keren and Marom, 2014). This model provides a well-controlled setting for studying impacts of external inputs during development, as addressed by previous studies in which activity was sampled along development in-vitro following sparse stimulation phases (Wagenaar et al., 2006; Bologna et al., 2010).

This work, extends these studies in the direction of continuous recording and stimulation of cortical networks in the course of development, as of the second day in-vitro. The experimental database is comprised of six networks developing while exposed to continuous stimulation, as well as an additional two non-stimulated networks, similarly monitored. Stimulation was applied alternately from two spatially distinct sources with a uniform distribution of inter-stimulus-intervals ranging from 6 to 14 s. Ultra-slow perfusion of culture medium was used to ensure this ongoing long-range experimental setting. Examples given in Figure 1, present extracts of recorded activities during the ninth day in-vitro of two non-stimulated networks (A) and two continuously stimulated ones (B). Note that in four of the six stimulated networks, parts of the sampled region could not be detected to a certain extent, possibly due to deficiencies in technical monitoring, thus are absent from synchronized activity (synchronization properties, however, are still apparent and measurable in these cases, as exemplified in Figure 1B, left).

We chose to address this data using synchronization as a system variable, relevant for neural network function and activity and driven by external inputs. The development of synchronization in particular, has been suggested as the underling process for the fine-tuning of mature, functional neural circuits (Yuste et al., 1992; Katz and Shatz, 1996; Zhang and Poo, 2001; Leinekugel et al., 2002; Corlew et al., 2004; Khazipov et al., 2004; Kilb et al., 2011). Our experiments outline the gradual emergence of network synchronization between days 6 and 9, as shown in Figure 2A. Moreover, in this context, functionality can be approached through the capacity to classify (i.e., discriminate between) different inputs, a primitive that stands at the basis of cognitive capabilities. The development of efficient classification between two stimulation sources, estimated by recruitment orders of spike representations, is described in Figure 2B (left), as is its possible dependence on synchronization rate (example given in Figure 2B, right). Exposure to stimulation during development, is suggested to impact networks' mature activity, as demonstrated by a potential increase in synchronous activity properties of relative amplitude and duration (evident in Figure 1B), as well as by an increase of active electrodes participation during synchronization (see Figure 2C, reflecting the count of spikes during synchronization by each electrode). It should be noted that each network was cultured in a separate batch, repeating the same procedure using different animals.

Fruitful directions for research might include questions as diverse as the dependence of mature propagation orders of neurons across the network on initial activity-connectivity profiles (e.g., we found that units generating early activities are recruited later across the network in mature synchrony propagation, reflecting early activation convergence to a “follower” type of activation) and aspects of activity rhythms occurring along a scale of hours to days. Experimental extensions toward full reciprocal interaction with the environment by closing the loop on a key measure that is relevant for function and that is affected by external inputs, could modify the experienced environment and enhance underlying alterations of activity occurring during development (Shahaf and Marom, 2001; Wagenaar and Potter, 2004; Wagenaar et al., 2005; Arsiero et al., 2007; Rolston et al., 2010; Keren and Marom, 2014). Overall, we believe that the provided database and proposed interpretations have the potential to contribute insights regarding environmental interaction with the course of neural development.

OO performed the experiments, initial analyses, and created the database, HK performed the analyses and wrote the manuscript.