Towards neuro-memory-chip: Imprinting multiple memories in cultured neural networks

We show that using local chemical stimulations it is possible to imprint persisting (days) multiple memories (collective modes of neuron firing) in the activity of cultured neural networks. Microdroplets of inhibitory antagonist are injected at a location selected based on real-time analysis of the recorded activity. The neurons at the stimulated locations turn into a focus for initiating synchronized bursting events (the collective modes) each with its own specific spatiotemporal pattern of neuron firing.

Figure 1

(Color online) System used for local chemical stimulations. A micropipet mounted on a micrometer stage (a) is used for local chemical stimulation at a specific selected location. In (b) we show the tip of the micropipet placed above one of the recording electrodes $30\mu \mathrm{m}$ in diameter (the bar is $200\mu \mathrm{m}$). To control the volume of the stimulating droplets, the micropipet is connected to a syringe that is mounted on a second micrometer stage.

Figure 2

(Color online) Network response to local chemical stimulations. (a) Raster plot representation of the network response to a stimulation of a single droplet. The time axis is divided into $10\mathrm{ms}$ bins. Each row is a binary bar-code representation of the activity of an individual neuron—a bar or 1 is plotted each time the neuron fires an action potential. The red (vertical) line indicates the time of injection. The yellow (horizontal) rmarker represents the recorded neuron closest to the location of the stimulation. As seen, one injection is sufficient to initiate a single synchronized bursting event with a distinct pattern of neuronal firing. The activity starts at the vicinity of the stimulated location and continues to propagate along a fixed path—a specific neuron fires second, another third, and so on [29, 30]. (b) Network activity rate before, during, and after a stimulation session. As seen, the activity rate of the network is elevated while the network is stimulated. Once the stimulation session is over the network returns to its previous activity rate.

Figure 3

(Color online) Imprinting new classes of SBEs. (a) The inter-SBE correlation matrix before stimulations. (b) The effect of the first session of stimulations and (c) after the second stimulation session. The matrices in the three cases are reordered using the dendrogram clustering algorithm. (d)–(f) Result of projecting the correlation matrices on the two leading principal vectors of the PCA clustering algorithm. As seen on the left, (a) and (d), before stimulation, only one type of SBE is exhibited by the network. In the middle, (b) and (e), a new class of SBEs appears (red/right) in addition to the previously existing one (blue/left) indicating that it has been created by the stimulations. On the right, (c) and (f), we see that a third class of SBEs (the green cluster/right) is created by the second set of stimulations without erasing the two existing ones. The second session was applied $24\mathrm{h}$ after the first one. To present the propagation patterns, we plot the propagation diagram for the three classes of SBEs (g)–(i). We define for each neuron $i$ the time it fires its first spike SBE no. $n$ as $T_n^i$. We then draw for each SBE a colored dot for each neuron to indicate its relative time of firing $T_n^i$ using the jet color code: blue (dark) for early firing and red (light) for later firing. The blue (dark) arrows in the middle and right panels indicate the locations where the stimulations were applied.

Figure 4

(Color online) Persistence of the collective modes. (a) Rate of appearance of the three classes of SBEs after the second stimulation session (from a different network than the one whose results were shown earlier). The original existing class of SBEs is marked blue (dark), the first imprinted class of SBEs red (medium darkness), and the second imprinted class green (light). Although the marked red class was imprinted $24\mathrm{h}$ before the green one it still persists for more than $72\mathrm{h}$. While the appearance rate of the green (light) class first goes down it goes up again, and so on. (b) Stability levels of each group of SBEs. Each line represents the level of correlations between each SBE (of a specific class) and the representative SBE of that class (an averaged one over all the SBEs that belong to the class). As seen, the correlation levels are significantly high (average correlation level of $0.65\pm 0.05$) and preserved for more than $40\mathrm{h}$. The results demonstrate the robustness of the specific spatiotemporal firing patterns of each of the imprinted memory templates. (c) A segment of the recorded sequence of SBEs colored according to their spatiotemporal patterns. As seen, the spontaneous and imprinted SBEs are exhibited by the network, mixed in their order of temporal appearance.