Coherence resonance in bursting neural networks

Synchronized neural bursts are one of the most noticeable dynamic features of neural networks, being essential for various phenomena in neuroscience, yet their complex dynamics are not well understood. With extrinsic electrical and optical manipulations on cultured neural networks, we demonstrate that the regularity (or randomness) of burst sequences is in many cases determined by a (few) low-dimensional attractor(s) working under strong neural noise. Moreover, there is an optimal level of noise strength at which the regularity of the interburst interval sequence becomes maximal—a phenomenon of coherence resonance. The experimental observations are successfully reproduced through computer simulations on a well-established neural network model, suggesting that the same phenomena may occur in many in vivo as well as in vitro neural networks.

Figure 1

Experimental setup for SB recording, stimulation, and analysis. (a) An incubator housing a custom-made MEA system and light-stimulation tool. (b) A close-up view of the MEA dish with a cultured neural network on its top. (c) A custom-built signal amplification and electrical stimulation system. The neural signals picked up by a $8\times 8$ square-grid array of recording electrodes (with respect to four reference electrodes) are amplified $10000$ times in two stages—with a homemade 64-channel preamplifier (gain: $20\times$ at 1 kHz) and a 64-channel filter amplifier (gain: $500\times$ at 1 kHz). The signals are sampled at 10 kHz (with 12-bit resolution) with an analog-to-digital converter (DAQ-2204, Adlink Technology) and stored and analyzed in a Linux computer. The light-stimulation system consisted of a blue light-emitting diode (LED) (emission peak at 460 nm, 150 $\mu \mathrm{m}$ in diameter, ASMT-JB31-NNP01, AVAGO Technology), a homemade precision LED current source, and an Arduino Uno board for computer interface and control. The LED light was focused onto the sample MEA dish through a condenser lens ($f=1$ cm) and an objective lens (long working distance, $35\times$, NA 0.45, Narishige).

Figure 2

Exemplary raster plots showing spontaneously generated synchronized neural bursts in (a) in vitro cortical neuronal culture and in (c) Izhikevich neural network model. Panels (b) and (d) are IBI histograms, based on a time series of 900 s and 100 s, corresponding to (a) and (c), respectively. A whole cortex dissected from a postnatal (within 3 days) Sprague-Dawley rat was dissociated and the dissociated cells were plated on a 64-channel, planar MEA dish (MED-P515A, Apha MED Sciences) and kept in an incubator (SCA-165D, Astec), which maintained 5% ${\mathrm{CO}}_2, 37{}^{\circ }\mathrm{C}$, and 100% humidity. The MEA dish was filled with 2 ml neurobasalmedium (21103-049, Gibco) with 2% B-27 supplement (1704-044, Gibco) and 1% L-glutamine (25030-081, Gibco) and the half of the medium was changed twice per week thereafter. See Ref. [22] for more details. The culture sample is DIV 55 at time zero. Only the peak positions of individual spikes are plotted in panels (a) and (c). For the Izhikevich model (see the text for details), the cell number from 0 to 159 (160 to 199) corresponds to an excitatory (inhibitory) cell. The burst positions are identified based on our custom algorithm as described in Refs. [22, 28, 34].

Figure 3

Effects of light stimulation on ChR2-expressing cells in a network: (a) some exemplary time traces of membrane potential, (b) mean membrane potential, and (c) membrane potential fluctuation (standard deviation) for different levels of light intensity. The time series shown in the second row of (a) is a close-up view of the boxed areas in the first row of (a). For the estimates given in (b) and (c), only the time domains (total 20 s long) that are devoid of APs are included. The error bars in (b) and (c) represent the degree of fluctuations in the control (light intensity 0) state roughly over a duration of 30 min. Due to the limited time of stable patch-clamping recording, only single measurement was done for other light intensity settings. The culture sample was grown on a coverslip at a density of $200\text{cells}/{\mathrm{mm}}^2$ with the same culture medium that was described previously. Adeno-associated virus stock solution [rAAV2/hSvn-hchR2(H134R)-eYFP-WPREpA, Vector Core at the University of North Carolina at Chapel Hill] was added at DIV 7. The recordings were done at DIV 21 in a typical whole-cell patch-clamp setup with oxygenated normal artificial cerebrospinal fluid perfusion in unclamped recording mode at room temperature as described in Ref. [41].

Figure 4

Coherence resonance of synchronized bursting in vitro neural network (a) $C_{\mathrm{var}}$ and $D_{\mathrm{eff}}$ as a function of the intensity of light stimulation in a cultured neural network (initial cell density: $1.4\times {10}^4{\text{cells/mm}}^2$). Adeno-associated virus stock solution [rAAV2/hSvn-hchR2(H134R)-eYFP-WPREpA] was added at DIV 21. A very similar coherence resonance phenomenon is realized with the Izhikevich model using overall offset current (b), the number of neurons receiving a noisy input current (c), and the amplitude of noise current (d), as a control parameter. Error bars in (a) reflect the variation over three runs taken over a time span of 106 days. Error bars in (b), (c), and (d) represent standard deviations based on five different runs made with different initial conditions.

Figure 5

Functional relationship between mean AP firing rate vs optogenetic light stimulation intensity in experiment (a), offset current (b), the number of neurons receiving noise input (c), and the amplitude of noise (d). The mean firing rate in (a) was evaluated based on the spiking activity recorded at 45 active channels over 1200 s, while those in (b)–(d) were were evaluated based on the activity of 160 excitatory neurons over 100 s.

Figure 6

Damped oscillatory response following a single pulse stimulation: (a) A raster plot of neural spikes and IBI time sequence immediately following a pulse stimulation delivered at time 0 in experiment (DIV 87). (b) A very similar result obtained by a computer simulation on the Izhikevich model. The graph in the right column of (a) is a close-up view of the blue boxed area containing a single SB. We should note that each electrode (channel) picks up signals only from its close vicinity, thus, it is quite possible that many single events of APs go unnoticed.