Phase-Dependent Effects of Stimuli Locked to Oscillatory Activity in Cultured Cortical Networks

Abstract

The archetypal activity pattern in cultures of dissociated neurons is spontaneous network-wide bursting. Bursts may interfere with controlled activation of synaptic plasticity, but can be suppressed by the application of stimuli at a sufficient rate. We sinusoidally modulated (4 Hz) the pulse rate of random background stimulation (RBS) and found that cultures were more active, burst less frequently, and expressed oscillatory activity. Next, we studied the effect of phase-locked tetani (four pulses, 200 s⁻¹) on network activity. Tetani were applied to one electrode at the peak or trough of mRBS stimulation. We found that when tetani were applied at the peak of modulated RBS (mRBS), a significant potentiation of poststimulus histograms (PSTHs) occurred. Conversely, tetani applied at the trough resulted in a small but insignificant depression of PSTHs. In addition to PSTHs, electrode-specific firing rate profiles within spontaneous bursts before and after mRBS were analyzed. Here, significant changes in firing rate profiles were found only for stimulation at the peak of mRBS. Our study shows that rhythmic activity in culture is possible, and that the network responds differentially to strong stimuli depending on the phase at which they are delivered. This suggests that plasticity mechanisms may be differentially accessible in an oscillatory state.

Introduction

Multi-electrode arrays (MEAs) offer a means to investigate synaptic plasticity on the small-network scale (1–3). Cultures of dissociated cortical neurons create a monolayer of cells on the MEA surface that are easily accessible for recoding and stimulation, and may facilitate learning and memory studies (4). The occurrence of spontaneous, synchronized bursting in these nonstructured networks has made it difficult to achieve consistent results on plasticity using dissociated cultures (5–10). Bursts of action potentials, characteristic for networks of dissociated neurons, resemble the type of activity that is observed during early development of the nervous system. This activity subsides as the brain starts to receive input from sensory neurons (11–14); thus, a lack of afferent external input to cultures may cause bursting. The fact that bursts can be suppressed by random background stimulation (RBS) supports this view (7). Given the high rate of action potential firings, and the fact that NMDA receptors are activated during bursts, it is reasonable to assume that plasticity mechanisms are activated in an uncontrolled manner (15,16). Furthermore, it has been shown that plasticity mechanisms are more accessible by stimuli when bursts are suppressed (17–19). Thus, the ability to suppress bursts in cortical cultures may be beneficial for accessing and assessing synaptic plasticity.

In this study, we took inspiration from the rhythmic activity observed in the hippocampus, which is known to be involved in long-term memorization (20–24) and learning tasks (22,25). In particular, applying a train of stimuli (four pulses, 200 s⁻¹) at the peak of hippocampal oscillation results in long-term potentiation (LTP), whereas the same train of stimuli applied at the trough results in long-term depression (LTD) of postsynaptic potentials (22,23). One hypothesis about the mechanism involved is that the oscillation of inhibitory neurons modulates the excitability of neurons that are postsynaptic to the stimulated neuron and consequently modulate the direction of change (25). Oscillatory activity, regardless of its origin, may thus have a profound effect on the way in which stimuli are processed by the network.

In this study, we modified the RBS algorithm by using stimuli that were Poisson-distributed in time (average of 10 stimuli per second) and delivered to a fully randomized sequence of locations. Inspired by the oscillatory dependency of plasticity in vivo, we sinusoidally (4 Hz) modulated the Poisson parameter. By doing so, we found that the application of rhythmically modulated RBS (mRBS) in culture serves two purposes. First, it suppresses bursts that would otherwise obscure induced plasticity, and evokes oscillating activity in the network. Second, it modulates excitability in the culture such that phase-locked trains of stimuli have effects that are phase-specific.

We assessed changes in the network using two different methods. First, we considered the response to probe stimuli. We found that changes in the magnitude of responses were pathway-specific and indeed depended on the phase at which the tetani were delivered. Second, we analyzed spontaneous bursting activity. Previous studies have shown that the spatiotemporal patterns of activity during bursts are stable over periods of several hours (26–28) but can also be changed by proper electrical stimuli (29–31).

Materials and Methods

Culturing

Cells were obtained from the neocortices of newborn Wistar rats. The cells were dissociated mechanically by trituration, and chemically by treatment with trypsin. MEAs produced by Multi Channel Systems (Reutlingen, Germany) were coated with poly-ethyl-imine to promote cell-to-surface adhesion. Next, a drop of plating fluid was placed in the center of the MEA and the cells were allowed to attach for 4 h, after which the plating medium was washed and nonadhering cells were removed. The plating concentration was 1 million cells per milliliter, which resulted in a monolayer of cells with a density of ∼2500 cells per mm² at 2 days in vitro (DIV). We used Romijn's chemically defined R12 medium (32) for plating, maintenance, and experimentation. Half of the medium was replaced every 2 days. Cultures were stored in an incubator at 37°C, 5% CO₂, and near 100% humidity. The platinum-nitride MEA electrodes were 30 μm in diameter and spaced 200 μm apart in an 8×8 grid (excluding the corner electrodes). The electrodes were numbered in matrix form (e.g., electrode 45 was located at the fourth column, fifth row). A total of eight different cultures, taken from five different platings, were used. Culture ages ranged from 20 to 89 DIV, with most (13/17) recording sessions performed between 20 and 35 DIV.

Experimental setup

Measurements

We built our measurement setup around a commercially available MEA-recording setup (1060BC preamplifier, STG1002 stimulus generator) from Multi Channel Systems. Data were acquired at a 16 kHz sampling rate with the use of a 6024E card (National Instruments, Austin, TX), and controlled by custom LabView programs (National Instruments). During measurements, the MEAs were sealed with a semipermeable membrane (Multi Channel Systems) and the temperature was controlled at 36°C. The CO₂ level was maintained at 5% and the culture chamber was heated from the top by a Peltier element to prevent condensation of the medium. Cultures were left to acclimatize in the experimental setup for 30 min before experiments were initiated. The stimuli were always monopolar, biphasic (200 μs per phase, positive phase first) current pulses. During mRBS, the same pulse was used for all electrodes.

During probing sessions and spontaneous recordings, spikes were detected using a threshold crossing algorithm and validated in real time using an algorithm described by Wagenaar et al. (33). During mRBS sessions, continuous traces of all 60 electrodes were saved because spike validation interfered with the timing accuracy of stimuli. These data files were analyzed offline in the same way as would otherwise have been done in real time.

Protocol

All 60 electrodes were randomly probed four times with three different amplitudes (between 8 and 20 μA) at 5 s intervals to determine the subset of electrodes to be used for mRBS. Ten to 13 electrodes whose poststimulus histograms (PSTHs; summed across all electrodes) showed a peak larger than baseline activity were chosen for further use. A single stimulus amplitude was selected for all electrodes and for the entire duration of the measurement. Once the electrodes had been chosen, the experiment consisted of evaluation periods interleaved with short periods of mRBS (Fig. 1). Each evaluation period included 1), a sequence in which a subset of electrodes were probed such that PSTHs could be calculated; 2), a period in which there was only spontaneous activity; and 3) another probing sequence. Probing sequences involved all electrodes that were also used for mRBS in fully randomized order, using fixed amplitudes at 5 s intervals and repeated 15 times.

Figure 1.

Timeline of a measurement. First, all electrodes were probed with stimuli of three different amplitudes. Based on the array-wide PSTHs, a selection of 10–13 electrodes was made for further use. Next, evaluation periods were interleaved with mRBS stimulation or sham periods of no stimulation. An evaluation period consisted of two probe sessions and 30 min of spontaneous recording. Both probe sessions were used, such that P1 and P2 were compared with P3 and P4 in the analysis. At least two different settings were used (0° and 180°; long arrows in insets indicate tetani) in random order.

mRBS stimulation

Stimuli were generated in a probabilistic manner in both time and space. At every step dt, one randomly selected electrode was allowed to be activated. The probability of activation (i.e., the rate) was time-varying, as:

$$r\left(t\right)=r_0+r_1\text{cos}\left({\omega }_{m}t\right)\text{.}$$ (1)

Where the angular velocity of modulation, ω_m = 2πf_m, was chosen such that f_m = 4 Hz. Random activation events with instantaneous probability r(t) were generated by comparing at each time step a pseudo-random number x, uniformly distributed between 0 and 1, to r(t) · dt. The resulting time series was Poisson-distributed with a time-varying parameter (i.e., r(t)). The average rate of stimulation, r₀, was set to 10 s⁻¹, which was enough to significantly suppress bursting (as discussed in Results; but see also Wagenaar et al. (7)). The maximal deviation (r₁) was set to either 8 or 10 s⁻¹. The time step dt was set as small as technically possible, which was determined at 20 ms. The sequence of stimulation sites (10–13 sites) was fully randomized. Consecutive stimuli on the same electrode were allowed.

Phase-locked tetani

At particular phases of the modulation frequency of mRBS, tetani consisting of four pulses at a rate of 200 s⁻¹ were applied to one electrode that was not already being used for RBS. All experiments consisted of at least three mRBS stimulation sessions with phase-locked tetani. The control group was formed by sham stimulation sessions in which no stimuli at all were applied. The parameters used are summarized in Table 1.

Table 1.

Parameters used for several modes of mRBS stimulation

Intervention	Phase	Pulses	ITI	No. of experiments	No. of cultures
mRBS (in phase)	0°	4	5 s	20	8
mRBS (out of phase)	180°	4	5 s	20	8
Control (no stimulation)	-	-	-	13	8

A sham stimulation session of equal duration was taken as control. Tetani were repeated every 5 s during 6 min of mRBS (see Fig. 1). The last two columns of Table 1 list the number of experiments and the number of cultures used.

We use cosine notation, such that 0° corresponds to peak mRBS stimulation rate (18 or 20 s⁻¹) and 180° to the minimal mRBS rate of 0 or 2 s⁻¹ (see Fig. 1). In addition, because of setup limitations, mRBS was suspended from 40 ms before tetanus to 40 ms after tetanus. The number of pulses per tetanus was chosen in accordance with Huerta and Lisman (22), who found that four pulses (at 200 s⁻¹) applied at the peak or trough of the hippocampal θ oscillation were enough to induce LTP or LTD (respectively) with the same magnitude as traditional protocols. Subsequent groups used four to five pulses at either 200 or 400 s⁻¹ (400 s⁻¹ chosen to limit tetanus duration) (21,23).

PSTH analysis

The network state was evaluated with the use of PSTHs made during evaluation periods. Electrode-specific PSTHs were made of responses to probe stimuli delivered through each electrode used also for mRBS. A PSTH for each electrode was calculated as the average histogram of activity at 0–500 ms after probe stimuli, using bins of 5 ms. The change in area of the PSTHs was taken as a measure for changes in network responses:

$${\Delta }_{i,j}=\frac{{\text{area}}_{i,j}^{\text{after}}-{\mathrm{area}}_{i,j}^{\text{before}}}{{\text{area}}_{i,j}^{\text{before}}}$$ (2)

Where i is for every electrode, j denotes stimulus electrodes, and “before” and “after” are with respect to an mRBS intervention (Fig. 1). We disregarded PSTHs with an area^before smaller than 10 spikes to decrease numeric fluctuations caused by division by a small number. This also vastly reduced the amount of PSTHs that merely contained direct responses (i.e., responses that involved no or very few synapses). A normalized histogram of changes (all Δ_i,j of all experiments) was made to visualize the aggregate results of the applied interventions. We tested whether histograms originated from two different distributions by using a two-sample, single-sided Kolmogorov-Smirnov test (α = 0.05). Another measure that is more sensitive to the extremes of histograms is the potentiation ratio (PR), which is calculated as the number of potentiated (Δ > 20%) connections divided by the number of depressed connections (Δ < 20%). The PR takes into account the fact that most connections in the network remain unchanged. We derived the PR from the work of Chiappalone et al. (34), who also proposed the threshold of 20%. Although the relation between PSTHs (as described here) and synaptic plasticity is far from direct, several previous studies revealed concomitant changes when MEA activity was measured in parallel with intracellular activity (5,35–38).

Burst profile analysis

In addition to PSTHs, we analyzed bursts during spontaneous sessions using a previously described method (26). In brief, we calculated the array-wide spiking rate by convolving the spike-train with a Gaussian with a resolution of 1 ms. Whenever the resulting signal crossed a threshold, a part of the signal centered to the first local maximum after threshold crossing was selected. We also calculated the spiking rates of individual electrodes by convolving spike trains with a Gaussian. Thresholds were adjusted to the culture's activity. The standard deviation used to calculate single electrode spiking rates was 5 ms. Selected episodes extended from 200 ms before, to 400 ms after the time of the maximum array-wide spiking rate. This was wide enough to capture the whole burst in all measurements. In accordance with earlier work, we call the Gaussian-smoothed, array-wide firing rate the “burst profile”, and the electrode-specific smoothed firing rate the “phase profile”. Before conducting further analyses, we calculated the averages of the profiles over 5 min bins. This reduced the variability and computational load. We treated these averaged profiles as if they were individual profiles. Changes in profiles were calculated as the magnitude of the difference:

$${\text{change}}_{i,j}={\Vert {\text{profile}}_j-{\text{profile}}_i\Vert }_{i\ne j}$$ (3)

where i and j enumerate bursts in a measurement, and profile_j is a 601-element (i.e., 600 ms) vector describing either a burst profile or a phase profile. To assess changes, we created two groups. The between-group consists of changes between profiles that are separated in time by mRBS intervention, and the within-group contains changes between profiles that are not separated in time by mRBS. The within-group represents natural variability of profiles, whereas the between-group represents variability plus a possible effect of intervention. We used Student's t-tests to determine significance (α = 0.05). For clarity of presentation, the between-group changes depicted in Fig. 6 were normalized to the mean within-group changes.

Results

mRBS

Fig. 2 gives an overview of mRBS combined with tetani applied at 0°. The average rate of stimulation (r₀) was 10 s⁻¹, modulated with a rhythm (f_m) of 4 Hz and deviation (r₁) of 8 s⁻¹. We observed that activity lagged stimulation by ∼13 ms. This corresponds to the typical response time of the culture to stimuli, which is also reflected in the typical PSTH. The interspike interval distribution of observed activity during mRBS strongly resembles that of the stimuli, with a relative increase at intervals near 0.25 s due to 4 Hz modulation. In contrast, the interspike intervals of spontaneous activity are more broadly distributed, with the longer lags (e.g., > 0.5 s) corresponding to the relatively quiet times between bursts. It is particularly interesting that the number of intervals smaller than 20 ms is ∼5 times lower during stimulation than during spontaneous activity, indicating less bursting. The right panels in Fig. 2 illustrate the burst suppression by mRBS. By investigating rhythmic activity instead of bursting activity, one can study plasticity in a more controlled manner.

Figure 2.

mRBS stimulation and activity. Top left panel: Histograms of stimuli (black) and activity (dashed) with tetani applied at 0°. Activity lagged stimuli by ∼13 ms and was scaled by a factor 0.2 for clarity. The activity was successfully modulated by 4 Hz mRBS stimulation. Lower left panel: Interevent interval distribution. Curves 1 and 2 show the interspike interval distributions during stimulation and spontaneous activity, respectively. Curve 3 shows the interstimulus interval distribution. Under stimulation, long quiescent periods disappeared, and intense short-latency firing decreased (by ∼80%). Right panels: Congregate spike rate, in bins of 0.25 s, for 1 min of activity without stimulation (top panel) and during mRBS stimulation (lower panel). Stimulation increased overall activity but also suppressed bursts and general variability in the overall spike rate.

PSTHs

The PSTHs showed two phases in the responses (Fig. 3). Many electrodes measured activity only at very short latencies (<10 ms). Such early responses were dominated by direct responses, which are antidromic or orthodromic action potentials resulting from activation of an axon (39). Activity at longer latencies involves synaptic transfers, and was usually part of activity patterns that resembled network bursts. In the majority of cases, the shapes of PSTHs before and after intervention were similar, whereas the area underneath the curve changed.

Figure 3.

Example of PSTH changes. PSTHs in response to probe stimuli delivered to electrode 17 (dark shaded) are shown before (gray) and after (black) in-phase (0°) mRBS stimulation. In the complete experiment, mRBS stimulation was applied through all shaded electrodes, and tetani were delivered through electrode 45 (hatched). Twenty-two PSTHs with an area over 10 (spikes) were considered responsive. Of these, the area of 13 PSTHs increased by at least 20%, whereas that of only two PSTHs decreased by at least 20%. The bin width used was 5 ms. Electrodes are numbered in column-row fashion.

In most experiments, a mix of increased and decreased PSTH areas was found, even when one probe electrode was considered, as shown in Fig. 3. The implication is that changes were pathway-specific rather than stimulus-site specific. Such an observation argues against the possibility that the excitability of the stimulated neuron was somehow changed. Rather, it suggests that the synaptic connections between stimulated and observed neurons were changed. Aggregate results of the changes in PSTH area of all individual electrodes that were observed for the two different mRBS settings and control experiments are shown in Fig. 4. When tetani were delivered in-phase (at 0°) of the modulation frequency, the PSTH areas increased significantly from those observed when no stimulation was applied (p = 0.047, single-sided, two-sample Kolmogorov-Smirnov test). Conversely, tetani delivered in antiphase (at 180°) did not result in a significant decrease in overall PSTH area (p = 0.061).

Figure 4.

Aggregate results for changes in PSTH area. Without stimulation, the dotted curve was found. The mean change in the absence of mRBS ($\overline{\Delta }$) was −3.1%, indicating an overall decrease of PSTH area. With tetani applied in-phase (0°) with mRBS, a shift toward increases in PSTH area was found ($\overline{\Delta }$: +5.7%; solid line). The shaded area indicates the standard deviation of the histogram. Slight decreases in PSTH area were found for tetani applied at 180° of mRBS ($\overline{\Delta }$: −6.2%). Bin width: 2.5%.

The overall trend was toward a decrease in PSTH area. This is best seen by dividing the area of the curve showing a >20% increase by the area of the curve showing a >20% decrease. The PR was 0.62 when no stimulation was applied, 0.39 for tetani applied at 180° of the modulated stimulation, and 1.84 for tetani applied at 0°.

Burst analysis

To examine the effect of our interventions on spontaneous activity, we recorded 30 min of bursting activity before and after each intervention (Fig. 1). The firing rates of individual neurons during a burst for one particular example are shown in Fig. 5. Although there is usually a short period in which all contributing electrodes synchronize, there are also differences between contributing electrodes. For example, electrode 13 was usually early to fire, whereas electrode 86 contributed mainly to the late phase of bursts. The specificity of electrode contributions (which we call phase profiles) is caused by the underlying network connectivity and pathways of activation. Changes in activity during bursts included a shift in the time of the main phase of firing, and more (or less) intense firing. In Fig. 5, the tendency was toward broader profiles and higher peak activity, indicating that the bursts had become more intense as a whole.

Figure 5.

Examples of phase profiles before (gray) and after (black) mRBS stimulation with tetani at 0°. The profiles are averages of the last 10 min of spontaneous recording before mRBS stimulation, and the first 10 min after. Spontaneous recordings from the same experiment as in Figs. 3 and 4 were used.

Overall, the analysis of bursts shows that in-phase tetani were able to induce significant changes in activity patterns, both in the aggregate firing rate (burst profiles, p = 0.040, two-sided Student's t-test) and in individual electrode contributions (phase profiles, p = 0.025). Tetani delivered in antiphase of the mRBS were unable to induce changes in either burst profiles (p = 0.182) or phase profiles (p = 0.134). In fact, there was a tendency toward stabilization of existing patterns, as the mean distance between profiles was slightly smaller when antiphasic stimulation was applied than in controls in which no stimulation was applied. Fig. 6 summarizes the statistical analysis of changes in spontaneous bursts.

Figure 6.

Changes in burst profiles (left panel) and phase profiles (right panel) for different settings of mRBS stimulation. The bars indicate the mean change after intervention, relative to inherent variability of profiles. The mean changes induced by mRBS stimulation are compared with ongoing changes in the absence of stimuli (control). All profiles from all experiments were included. Error bars denote standard error of mean; ^∗ p < 0.05, two-sided Student's t-test.

Discussion

In this study, we investigated whether rhythmic activity in culture could produce phase-dependent plasticity. To this end, we first required that naturally occurring network bursts be replaced by rhythmic activity. This was successfully achieved by modulated and stochastic RBS, which decreased the percentage of small, interspike intervals that would normally be found in network bursts, and at the same time increased overall activity. These small interspike intervals are most likely to induce a spike-timing-dependent synaptic plasticity that may otherwise obscure or negate experimentally induced changes in connectivity (4).

The PSTH analysis showed a general decrease in responsiveness for the no-stimulation controls, which is not uncommon in regularly stimulated cultured neuronal networks (40,41). However, when tetani were applied at 0°, an increase in overall PSTH area was observed. The decrease in PSTH area by tetani applied at 180° was not significant, but it may have been partially obscured by the general decrease of PSTH area found in controls. Nevertheless, applying mRBS with phase-locked tetani had an effect in networks of dissociated neocortical neurons, which depended on the phase at which the tetani were applied. The burst analysis showed results similar to those obtained by PSTHs, with only tetani applied in-phase with mRBS resulting in a significant increase in distances between profiles as compared to controls. We did not test the duration of these changes, but they extended at least 1 h after intervention. In this study, recordings of spontaneous activity started 15 min after intervention and were stopped 15 min before the next intervention. Therefore, the results from the burst analysis may have been negatively influenced, especially when one considers that probe stimuli were delivered during these two blocks of 15 min, which may alter activity patterns by themselves (31).

There may be another explanation for the reduced effect of synaptic depression. It is possible that the average synaptic efficacy is relatively low, and thus there is less opportunity for a further decrease. Since most studies have focused on potentiation using extracellular recordings, there is no evidence to support this. However, Jimbo et al. (5,36) confirmed a PSTH decrease by whole-cell patch-clamp recordings, demonstrating that depression is possible and that PSTHs are sensitive to depression.

Overall, our results indicate that tetani applied in-phase with mRBS had a definite effect, whereas the effect of antiphasic tetani was much smaller at best. This difference in effect suggests the importance of the network-wide firing rate at the moment the tetani are applied. The efficacy of tetani may be time-varying because mRBS modulates the firing threshold for neurons that are postsynaptic to the stimulus site. Because stimuli were generated at a relatively low average rate in a probabilistic way, and mRBS was discontinued from 40 ms before tetanus to 40 ms afterward, it is unlikely that such scaling of excitability is a direct result of stimulation. Rather, mRBS may excite neurons to oscillate at an intrinsic oscillation frequency near the modulation frequency. Cortical neurons have a resonance frequency near 4 Hz (42,43). In this respect, it is interesting that Wagenaar et al. (9,10) previously combined RBS (fixed aggregate rate of 50 s⁻¹, cyclic electrode switching) with tetanic stimulation (20 trains, 20 pulses/train, 20 s⁻¹, 2 s between trains) but found no significant changes in responses to test stimuli. Our study shows that rhythmic activity and the phase of tetanus delivery are important factors that strongly influence further processing in a neuronal network.