INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Throughout the developing nervous system, immature neural circuits spontaneously generate periodic activity patterns (Yuste 1997). There is growing evidence that this spontaneous correlated activity can play a critical role in the formation of adult networks by influencing differentiation (Berridge 1998; Spitzer et al. 2000), motility (Gomez and Spitzer 1999; Lautermilch and Spitzer 2000), and connectivity (Ben-Ari et al. 1997; Katz and Shatz 1996; Wong 1993). In the developing retina, periodic bursts of action potentials propagate across the ganglion cell layer (Meister et al. 1991), influencing the development of both circuits within the retina (Bansal et al. 2000; Sernagor and Grzywacz 1996; Sernagor et al. 2001) and retinal projections to central targets (Wong 1999).

The mechanisms underlying the spontaneous generation of periodic activity have been studied in great detail in the developing hippocampus, spinal cord, and retina. Although the specific architecture of these three circuits differs significantly, qualitatively the activity across these regions is highly similar (Feller 1999; O'Donovan 1999). Namely, spontaneous bursts of action potentials are correlated across a network of highly connected neurons causing periodic increases in intracellular calcium concentration ([Ca²⁺]_i). These patterns are generated solely via excitatory neurotransmission. This stands in sharp contrast to "classic" periodic networks whose rhythmicity is created by reciprocal excitatory and inhibitory connections that function as a pacemaker unit (reviewed in Harris-Warrick et al. 1992) or by time variant conductances, such as hyperpolarization-induced current (I_h), that are involved in membrane oscillations (Angstadt and Calabrese 1989; Luthi and McCormick 1998; Pape 1996; Thoby-Brisson et al. 2000). How does a network of neurons maintain periodic activity without the aid of pacemaker circuits or conductances?

Here, we show that immature mammalian retinal neurons in culture form synaptically coupled networks that undergo periodic activity. This activity, measured as periodic compound postsynaptic currents (PSCs) and [Ca²⁺]_i transients, is correlated over large distances and is driven primarily by glutamatergic synaptic transmission between retinal ganglion cells (RGCs). We also find that evoked responses at ganglion cell-ganglion cell synapses undergo synaptic depression, and the kinetics of this recovery from synaptic depression are similar to the interval between spontaneous events. Our results provide insight into strategies used by excitatory networks to generate correlated periodic activity.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Preparation of retinal cultures

Cultures containing neurons and glia were prepared from postnatal day 2 ferret tissue with a method that promotes the survival of RGCs (Meyer-Franke et al. 1995; Pfrieger and Barres 1997). Briefly, retinas were isolated in PBS containing gentamicin (Gibco) and were placed in Earles basic salt solution containing 15 U/ml of papain (PAP2; Worthington) and 0.04% DNase (Boehringer Mannheim). After tissue was rocked for 30 min, the supernatant was removed and replaced with an inactivating solution of trypsin inhibitor (1 mg/ml), 0.04% DNase, and ovomucoid (1 mg/ml) in neurobasal medium (Gibco). The cells were then dissociated by gentle trituration through a 1-ml pipette. The cell suspension was spun in a centrifuge (Beckman) at 800 rpm for 7 min, and the pellet was resuspended in a serum-free medium containing neurobasal medium, B27 additives (Gibco) glutamine, forskolin (50 µM), insulin, brain-derived neurotrophic factor (2 ng/ml), ciliary neurodrophic factor (CNTF) (200 pg/ml), fibroblast-derived growth factor (Peprotech), and gentamicin. Cells were plated on glass substrates treated with 10 µg/ml poly-D-lysine (70 kd; Sigma), and merosin or laminin (2 µg/ml) and were maintained in a humidified CO₂-O₂ incubator at 36.5°C. For electrophysiology measurements, cells were plated at a high density of 1.0-2.0 × 10⁵ cells/cm². For routine immunohistochemistry, cells were plated at one-fifth this density.

Immunohistochemistry

FIXED-CELL STAIN. Cultures were fixed in 4% paraformaldehyde in PBS for 10 min at room temperature and were rinsed in PBS. Cells were maintained at room temperature while they were permeabilized in 0.4% Triton X for 30 min, rinsed three times in PBS, blocked in PBS containing 5% sucrose and 20% donkey serum for 1 h, and rinsed. They were incubated with primary antibodies (see below) overnight at 4°C. They were then rinsed, blocked again as above, and incubated in secondary antibodies (donkey anti-rabbit, anti-goat or, anti-mouse conjugated to either Rhodamine Red X or Cy3 (1:100; Jackson) for 60 min at 4°C. The primary antibodies used were polyclonal antibody markers for rabbit -aminobutyric acid (GABA; 1:2,000), rabbit glutamate decarboxylase (GAD-67; 1:2,000; Chemicon AB 108), and goat choline acetyltransferase (ChAT; 1:250; Chemicon) and monoclonal markers for rat and mouse Thy1.1 (1:80; Chemicon). In the case of Thy1.1, a surface marker, the Triton permeabilizing step was omitted.

LIVE CELL STAIN. Live cell staining was performed with a Thy1.1 antibody to identify RGCs (Taschenberger and Grantyn 1995). The primary antibody was added directly to the culture medium (1:30), and the cells were incubated for 45 min at 37°C. The cells were rinsed, incubated with a FITC-conjugated phycoerthyrin anti-mouse secondary antibody (1:50; Jackson) for 30 min, and fixed in 4% paraformaldehyde in PBS for 15 min.

Electrophysiological recordings

Whole cell patch-clamp recordings were made, either in voltage-clamp or current-clamp mode, with an Axopatch 200B patch-clamp amplifier (Axon Instruments). Voltage-clamp experiments were conducted at holding voltages of 60 mV. The intracellular solution for the current-clamp experiments contained the following (in mM): 98.3 K-gluconate, 1.7 KCl, 0.6 EGTA, 5 MgCl₂, 40.0 HEPES, 2 ATP-Na, and 0.3 GTP-Na, adjusted to pH 7.25 with KOH. The intracellular solution for the voltage-clamp experiments contained the following (in mM): 100 Cs-gluconate, 1.7 CsCl, 10 EGTA, 5 MgCl₂, 40 HEPES, 1 QX-314, 2 ATP-Na, and 0.3 GTP-Na, adjusted to pH 7.25 with CsOH. The external solution contained the following (in mM): 123 NaCl, 5 KCl, 3 CaCl₂, 2 MgCl₂, 10 D-glucose, and 10 HEPES, adjusted to pH 7.3 with NaOH. E_Cl was 60 mV for the above solutions. Electrode resistances varied from 2 to 5 MOhm. All recordings were performed at room temperature with perfusion rates near 1 ml/min.

All the recordings included in the analysis came from RGCs. Neurons that had large somas, significant Na⁺ current, and that fired repeated action potentials in response to current steps were presumed to be RGCs (Guenther et al. 1994). Neurons that had medium to large somas, small or no Na⁺ current, and fired a single action potential or no action potential in response to step depolarization were presumed to be amacrine cells (Bieda and Copenhagen 1999; Zhou and Fain 1996).

Synaptic currents were analyzed with pClamp6 (Axon Instruments) and Minianalysis 4.3.2 (Synaptosoft) software. To determine the frequency of the periodic synaptic currents, we constructed power spectrum densities, typically using 100 s of data (sample frequency 2.5 kHz). To allow easier computation without altering the overall structure of the data, we first interpolated data files to 2.0 × 10⁵ points.

For the synaptic depression experiments, a standard recording electrode was filled with external solution and was used for presynaptic stimulation. Stimuli (0.5-5 mA) were delivered for 50 µs at varying intervals (Iso-flex and Master 8, AMPI.). Recovery from paired-pulse depression was measured by monitoring the amplitude of evoked synaptic responses as a function of the interval between the stimulating pulses. Responses from different cell pairs were recorded at eight time intervals between 50 and 5000 ms; these were averaged, and the averages were fit to a single exponential. The error associated with the fit was estimated with an algorithm in IGORPro (Wavemetrics) that computes the SD of the fit parameters from the residuals and is based on the assumption that the residuals follow a normal distribution.

Optical recording

Cultures ranging from 15 to 70 days old were loaded with either fluo-3 AM or fluo-4 AM (Molecular Probes). Fifteen microliters of a 1 µg/µl solution of fluo-3 AM or -4 AM in DMSO containing 2% pluronic acid were added to 3 ml of culture media for 15 min at room temperature. After loading, the cultures were washed and perfused with external solution. Intracellular fluorophores were excited at 480 nm and imaged with either a 10× (Zeiss CP-Achromat) or 40× (Zeiss Fluar) objective. Images were captured with a CCD camera (RTD-CCD-1300; Princeton Instruments) and were typically obtained with 100-ms exposures at 1 Hz.

Fluorescence images were obtained and analyzed with MetaMorph (Universal Imaging) and IGORPro. The fractional change in fluorescence (F/F) of the transients was calculated with the equation F/F = (F_t  F_base)/F_base, where F_t and F_base are the raw and baseline fluorescence values, respectively, obtained by averaging the fluorescence over the specified region of interest (ROI). For high-magnification imaging (40× objective), ROIs were defined for the soma of individual neurons; for low-magnification experiments (10× objective), ROIs were defined as 60 × 60-µm squares that encompassed roughly four neurons. No corrections were made for background fluorescence or bleaching because these factors were negligible. Typical F/F measurements ranged from 10 to 100%.

Plots of F/F obtained from the low-magnification images were used to construct raster plots (see Fig. 2D). Measurements of F/F versus time were computed for each ROI (averaged over 20 × 20 µm containing ~4 cells), and the second derivative was calculated. If the second derivative was above a threshold determined by eye (typically 2.5 to 5), a vertical line was drawn in the raster plot indicating a fluorescence transient. ROIs that did not contain transients were not included in the raster plot.

To determine the frequency of the [Ca²⁺]_i transients, we constructed power spectrum densities, typically using 100 s of data (sample frequency 1 Hz). See Electrophysiological recording for the procedure.

The effects of various pharmacological agents on the amplitude and frequency of [Ca²⁺]_i transients were averaged across 3-10 cells/culture, with the results normalized to the control values for that culture. These normalized results were then averaged across cultures, and, unless otherwise noted, the significance of the pharmacological effects was assessed by a two-sided t-test.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Retinal cultures contain predominantly ganglion and amacrine cells

Dissociated retinal cells isolated from postnatal day 2 ferrets were cultured for 15-70 days at high density. In the first few days after plating, cells extended processes that contacted neighboring cells. As the cultures matured, clusters of cells formed with a large concentration of glia (Fig. 1). Glial cells were retained in the cultures because they promote synapse formation (Meyer-Franke et al. 1995; Pfrieger and Barres 1997) and may support periodic circuit behavior (Verderio et al. 1999).

View larger version (146K): [in this window] [in a new window]   Fig. 1. Characterization of cell types in retinal cultures. Bottom: dissociated retinal cells grown in serum-free medium for 22-28 days were immunostained for Thy1.1, glutamate decarboxylase-67 (GAD-67) and choline-acetyltransferase (ChAT) and visualized with fluorescent secondaries. Top: corresponding phase images show large clusters of cells that contain primarily glia. Scale bar, 10 µm.

Immunohistochemical staining identified the cell types present in the cultures. Labeling with antibodies against the neuron-specific marker TeTx showed that a small percentage of cells in the culture were neurons (10.6 ± 2.91%; n = 7 culture preparations; 14-97 days in culture). The RGC-specific marker Thy1.1 identified the RGCs with staining in their somas and throughout their processes (Taschenberger and Grantyn 1995; Fig. 1A, left). Morphologically, the RGCs displayed larger soma sizes than did the other cell types, thus allowing their identification in the electrophysiology experiments. Antibodies against ChAT and the GABA-synthesizing enzyme GAD-67 identified cholinergic and GABAergic cells, respectively (Fig. 1A, middle and right). The cholinergic cells were, presumably, amacrines because a subset of this class provides the sole source of ACh in the retina (Famiglietti 1991; Kolb 1997; Tauchi and Masland 1984). The two main classes of GABAergic cells are amacrine and horizontal cells (Freed 1992; Massey and Redburn 1987). A horizontal-cell specific marker was not used to distinguish between these two cell types; only a few horizontal cells were seen, as identified by their characteristic morphology (Akagawa and Barnstable 1986). Accurately quantifying the percentage of neurons labeled for Thy1.1, ChAT, and GAD-67 was difficult because the density of neuronal cells present in the cultures was low. However, the predominant neuronal types in our cultures were amacrine cells and RGCs. All recordings described in the following experiments come from RGCs (see METHODS).

Cultured retinal neurons undergo periodic changes in [Ca²⁺]_i

We loaded our cultures with the calcium indicators fluo-3 AM or fluo-4 AM and monitored the spontaneous activity patterns by imaging changes in [Ca²⁺]_i. Using short loading times, we found the contribution from glial cells was negligible. Images were captured with either a high-power (40×; Fig. 2A, top) or low-power (10×; Fig. 2A, bottom) objective and were typically obtained with 100-ms exposures at 1 Hz. The fractional change in fluorescence (F/F) was plotted against time to show rhythmic fluorescence transients as seen in Fig. 2B (see METHODS). Plots of F/F obtained from low-magnification images were used to construct raster plots, in which each line signified a [Ca²⁺]_i transient for a local region of interest (Fig. 2D; see METHODS).

View larger version (83K): [in this window] [in a new window]   Fig. 2. Cultured retinal neurons display correlated rhythmic changes in [Ca2+]i. A: fluorescence (F) image of cultured retinal neurons loaded with fluo-3 AM visualized with 40× objective. Scale bar, 10 µm. B: example of F/F trace averaged over a single soma at 26 and 49 days in culture (DIC), recorded with 100-ms exposures at 1-s intervals. *, F/F transients caused by propagating waves. C: fluorescence image (10× objective) of cultured retinal neurons loaded with fluo-3 AM. Scale bar, 100 µm. D: raster plots mark [Ca2+]i transients averaged over neighboring 60 × 60-µm regions of the culture. Images were acquired with 100-ms exposures at 1-s (top) or 250-ms (bottom) intervals.

Based on the spatiotemporal properties of the measured spontaneous Ca²⁺ transients, we categorized the cultures into two groups: those that were 25-45 days in culture (DIC) and those that were 49-70 DIC. Cultures between 25 and 45 DIC displayed transient increases in fluo-3 fluorescence resulting from increases in [Ca²⁺]_i (16.2 ± 12.8% of the labeled cells; n = 50 cultures; Fig. 2, A and B). These [Ca²⁺]_i transients were highly regular, and a power spectrum analysis revealed a rhythm frequency that was uniform within a culture and that ranged from 0.10 to 0.60 Hz across cultures (corresponding roughly to interevent intervals between 2 and 10 s.). In these cultures, [Ca²⁺]_i transients were synchronized across cells that were separated by over 1 mm (Fig. 2, C and D; for Quicktime movie, see www.biology.ucsd.edu/labs/feller/JNeurophys.html, clip1). These transients may have resulted from activity propagating >1 mm/s (Maeda et al. 1995) but appeared synchronous in our recordings because of our limited time resolution.

In contrast to these young cultures, cultures over 49 DIC displayed [Ca²⁺]_i transients that propagated across the cells (n = 11 cultures). This propagation is shown by a raster plot constructed from images acquired at 4 Hz (Fig. 2D, bottom). The propagating transients originated at irregular time intervals and multiple locations, suggesting that there were no "pacemaker" clusters of cells that were responsible for wave initiation (for Quicktime movie, see www.biology.ucsd.edu/labs/feller/JNeurophys.html, clip2). Moreover, all regions examined (n = 9 fields of view in each culture) contained cells that were capable of supporting propagating activity. Waves propagated at a speed of 200-500 µm/s, although analysis of the propagation speed was difficult because wavefronts were not clearly defined (Feller et al. 1997).

Periodic activity is driven primarily by glutamatergic synaptic transmission between RGCs

To determine if Ca²⁺ transients in retinal cultures were driven by spontaneous synaptic activity, we performed whole cell recordings of RGCs. In cultures over 21 DIC, RGCs revealed robust synaptic activity. A subset of RGCs (37/137 cells, 11 cultures) displayed periodic compound PSCs manifested as a sustained (0.5-1 s) compound burst of inward currents terminated by a period of sparse activity (Fig. 3A). Power spectrum analysis identified a single prominent frequency component for each culture (Fig. 3B). Across cultures, the prominent frequency component ranged from 0.10 to 0.60 Hz, comparable with the frequency measured for spontaneous Ca²⁺ transients. The range in peak frequency did not correlate with the age of the culture. The remaining RGCs (100/137) were classified as nonperiodic cells because they displayed no peak in the power spectrum. However a subset (27/100) of these nonperiodic cells did display compound PSCs.

View larger version (17K): [in this window] [in a new window]   Fig. 3. Spontaneous periodic compound postsynaptic currents (PSC)s and postsynaptic potentials (PSPs) measured in cultured neurons. A: continuous whole cell voltage-clamp recordings of a retinal ganglion cell (RGC). Cell was held at 60 mV. Top: expansion of a single compound PSC. B: power spectrum density computed from a whole cell voltage-clamp trace. The spectral peak for the cell was 0.195 Hz. C: continuous whole cell current-clamp recording from a RGC resting at 50 mV. The action potential peaks are truncated and shown at top.

In many preparations, the periodicity of neural activity has been shown to involve a time-variant intrinsic cellular conductance, such as the hyperpolarizing activated current termed I_h (Angstadt and Calabrese 1989; Luthi and McCormick 1998; Pape 1996; Thoby-Brisson et al. 2000). To determine if I_h or any other intrinsic conductance was involved in our regular activity, we performed whole cell current-clamp recordings (Fig. 3C). We found that RGCs displayed no changes in baseline membrane voltage between depolarizations (n = 15 cells), indicating that no time-varying intrinsic conductance was activated. In addition, bath application of artificial cerebrospinal fluid (ACSF) containing 2 mM Cs⁺, a potent antagonist of I_h, did not prevent periodic depolarizations but did reduce the frequency of compound PSCs (frequency_Cs/frequency_control = 0.46 ± 0.06; n = 5). Given the effects of Cs⁺ on other conductances, it is difficult to interpret the change in frequency. However, these results indicate that I_h is not required for the generation of periodic activity.

We also investigated the source of the compound PSCs measured in RGCs. Periodic events >30 pA, which constituted all of the periodic compound PSCs, were blocked by Cd²⁺ (n = 6), whereas spontaneous miniature events <10 pA that were nonperiodic, were unaltered. 6,7-Dinitro-quinoxaline (DNQX) abolished all synchronous periodic Ca²⁺ transients (n = 8 cultures; Fig. 4A) and periodic compound PSCs (n = 6 cells; Fig. 4B), indicating that the periodic events were driven by glutamatergic transmission. TTX, a blocker of voltage-gated Na⁺ channels, also abolished these periodic Ca²⁺ transients (n = 5 cultures) and compound PSCs (n = 7 cells), indicating that transmission comes from other RGCs, because they are the primary retinal cell type with Na⁺ action potential-based glutamatergic release. Together, these observations are consistent with the idea of compound PSCs of RGCs being generated by glutamatergic release from other RGCs.

View larger version (24K): [in this window] [in a new window]   Fig. 4. Rhythmic activity is blocked by inhibitors of synaptic transmission between RGCs. A: F/F averaged over 4 cell bodies located >50 µm apart, taken with 100-ms exposures at 1-s intervals. Bar, bath application of 25 µM 6,7-dinitro-quinoxaline (DNQX). B: whole cell voltage-clamp recordings from RGCs in control solution and solution containing 25 µM DNQX (middle). Spontaneous compound PSCs recorded in a RGC held at +20 and 50 mV in the absence and presence of 100 µM picrotoxin (PTX; bottom).

Because some neurons in the cultures were GABAergic (Fig. 1), we next investigated if GABAergic neurotransmission is involved in regulating or generating the periodic activity. Bath application of picrotoxin, which blocks GABA_A receptor-mediated Cl currents, blocked all outward PSCs recorded at +20 mV but had little effect on the inward PSCs recorded at 50 mV in the same cell (n = 7 cells; Fig. 4B, bottom). Changes in PSC amplitude and duration were not consistent across cells. Bath application of picrotoxin also had little effect on the periodicity of the spontaneous activity. The frequency of [Ca²⁺]_i transients in picrotoxin was slightly lower than that seen in control ([Ca²⁺]_i: frequency_picrotoxin/frequency_control = 0.77 ± 0.14; P < 0.05; n = 5). However, the interval between compound PSCs was not significantly affected (compound PSC: frequency_picrotoxin/frequency_control = 1.06 ± 0.431; P > 0.5; n = 5). Moreover, picrotoxin did not alter the amplitude of [Ca²⁺]_i transients (F/F_picrotoxin/F/F_control = 1.00 ± 0.15; P > 0.5; n = 6 cells) or the amplitude of the synaptic currents recorded when an individual RGC was held at 50 mV (PSC_picrotoxin/PSC_control = 0.97 ± 0.14; P > 0.5; n = 5 cells). These results indicate that GABAergic cells, although periodically active, have only a mild effect on the frequency and amplitude of periodic events in RGCs (Bacci et al. 1999).

Cholinergic synaptic transmission is required for spontaneous generation of correlation activity in the intact retina (Bansal et al. 2000; Feller et al. 1996; Sernagor and Grzywacz 1999). Because a subset of cultured neurons expressed ChAT (Fig. 1), we investigated whether cholinergic amacrine cells also participate in the periodic events observed in our cultures. Bath application of the nicotinic ACh receptor antagonists dihydro--erythroidine (DBE; 1 µM) or D-tubocurarine (curare, 100 µM) had variable effects on the cultures. In three of nine cultures tested, curare blocked all [Ca²⁺]_i transients; in the remaining six cultures, curare did not reduce the amplitude or frequency significantly [F/F_curare/F/F_control = 0.80 ± 0.33; P > 0.2; n = 6 and frequency_curare/frequency_control = 0.87 ± 0.13; P > 0.05; n = 5]. Consistent with this finding, the bath application of cholinergic antagonists did not alter the amplitude of synaptic currents (PSC_DBE/PSC_control = 0.99 ± 0.24; P > 0.5; n = 6 cells). These results indicate that cholinergic amacrine cells had a variable contribution to the spontaneous activity but, in most cultures studied, were not required for the generation of periodic activity.

In summary, our results indicate that periodic activity is mediated primarily by a glutamatergic, synaptically coupled RGC network although, in some cultures, cholinergic amacrine cells can contribute substantially. Although GABAergic transmission correlates with glutamatergic activity, GABAergic transmission does not influence the temporal pattern of glutamate release or the size of the [Ca²⁺]_i increases in retinal neurons.

Recovery from synaptic depression is similar to the periodicity of spontaneous activity

In the hippocampus (Staley et al. 1998) and spinal cord (Fedirchuk et al. 1999; Tabak et al. 2000), synaptic depression plays a critical role in generating rhythmic activity in vivo. To test if this mechanism might be involved in generating periodic activity in our retinal cultures, we examined paired-pulse synaptic depression. An extracellular pipette was used to stimulate a single visualized RGC to release glutamate to a nearby postsynaptic cell that was in whole cell configuration. This elicited inward PSCs (Fig. 5A) that were blocked by TTX (100 nM; n = 5 cells) or by DNQX (25 µM; n = 5 cells; data not shown). When two stimuli were delivered at intervals of <5 s, the second response showed depression (Fig. 5A, inset). Using pairs of stimuli with increasing interstimulus intervals, we found the time course of recovery from depression, as determined by a single exponential fit, to have a  of 2.0 ± 0.3 s (Fig. 5A), similar to that seen in several other synapses (for recent examples, see Dittman and Regehr 1998; Parker 2000; Staley et al. 1998; Tsodyks and Markram 1997). This recovery time is similar to the intervals between periodic compound PSCs recorded in these same cultures (2.0 ± 0.3 s; n = 6 cells; Fig. 5B). This result is consistent with the hypothesis that the kinetics of recovery from synaptic depression set the frequency of the periodic cells. However, to establish a causal relationship, additional experiments that test whether altering the recovery from synaptic depression affects the periodicity of the synaptic activity of the entire network are required.

View larger version (24K): [in this window] [in a new window]   Fig. 5. Time course of recovery from synaptic depression correlates with the frequency of rhythmic events. A: time course of recovery from synaptic depression is measured as the ratio of the amplitudes of the second PSC to the first (%PPD =1  B/A) averaged over all experiments. Solid line, single exponential fit; error bars, SD. Inset: whole cell voltage-clamp recording of pairs of PSCs evoked by extracellular stimulation of a neighboring RGC. Cells were held at 50 mV. B: continuous whole cell voltage-clamp recording from a RGC. Inset: power spectrum density for the cell.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

In this study, neural network behavior was examined in dissociated cell cultures generated from mammalian retinas. These retinal cultures consisted of periodically active networks whose pattern evolved with the age of the culture. In young cultures, imaging revealed near-synchronous [Ca²⁺]_i transients in cells that were correlated across much of the culture. In contrast, older cultures exhibited [Ca²⁺]_i transients that propagated across many cells at velocities similar to the "waves" seen in the intact developing retina (Wong 1999). Our study focused on the younger cultures in an effort to determine the cellular mechanisms underlying the periodic synchronous activity. These cultures contained periodically active networks that were synaptically coupled via excitatory glutamatergic and cholinergic transmission and whose periodicity was similar to the recovery kinetics from synaptic depression.

Comparison with in vivo circuits

Spontaneous periodic activity correlated across neighboring cells, either through mechanisms that lead to synchronous activation or to propagation, can be found throughout the developing nervous system (for reviews, see Ben-Ari et al. 1997; Feller 1999; O'Donovan 1999; Yuste 1997). In the developing mammalian retina, rhythmic trains of action potentials propagate across the RGC layer, causing periodic increases in [Ca²⁺]_i. Here we compare the circuit responsible for generating correlated activity in the culture dish to that of the intact developing retina.

Spontaneous activity in culture consists of periodic increases in intracellular calcium driven by barrages of excitatory synaptic input from a variety of cell types. Periodic depolarizations and compound PSCs are TTX- and DNQX-dependent mechanisms, indicating that glutamatergic transmission between RGCs is the primary source of excitatory coupling. In some cultures, cholinergic amacrine cells also provide a significant percentage of the excitatory input. In addition, GABA_A receptor-mediated currents contribute to the compound PSCs recorded in culture; however, blocking GABA_A receptors did not alter the periodicity of spontaneous activity. Hence, although GABAergic interneurons participate in the spontaneously active network, their input is not required for the generation of periodic activity.

In the intact developing retina, RGCs fire periodic propagating bursts of action potentials with a periodicity on the order of 1-2 min. This spontaneous activity is mediated by synaptic transmission between retinal interneurons, either cholinergic amacrine cells or bipolar cells, and RGCs. Retinal interneurons are known to release transmitter in a TTX-independent, graded-release manner. In contrast to the activity observed in culture, waves persist in the presence of TTX (Stellwagen et al. 1999) and even in the absence of RGCs (Stellwagen et al. 2000). The details of the circuit underlying the spontaneous generation of correlated activity in the intact retina changes with development. Early on, cholinergic synaptic transmission between amacrines and RGCs is required for the waves (Feller et al. 1996). Later, bipolar cell synaptic input to ganglion cells plays more of a predominant role (Bansal et al. 2000; Wong et al. 2000; Zhou and Zhao 2000). Although there is a substantial amount of GABA_A receptor-mediated current recorded during waves (Feller et al. 1996), blockade of GABA_A receptors does not affect the spatiotemporal properties of retinal waves (e.g., wave propation speed, size, and periodicity) during the period of development when retinal waves are mediated by activation of nAChRs (Stellwagen et al. 1999), similar to the role of GABA_A receptors in culture. However, at older ages in the intact retina, blockade of GABA_A receptors will significantly alter the observed firing patterns (Wong et al. 2000).

Although the circuit produced in the culture dish differs from that in the intact mammalian retina, the observation that developing RGCs form periodically active networks allows us to use an in vitro model to study the mechanisms that may underlie the periodic propagating activity patterns seen in vivo. Our results are consistent with the hypothesis that these correlated activity patterns emerge from a network containing excitatory synaptic connections. One hypothesis we have explored in culture is whether the periodicity of the activity is determined by synaptic depression.

Synaptic depression as a rhythm-generating mechanism

Rhythm generation is a prominent feature of the nervous system. Rhythmic motor behaviors such as walking, feeding, swimming, and respiration involve central pattern generators. In contrast, the periodic activity seen in dissociated cultures does not require the existence of inhibitory synapses or intrinsic conductances that cause membrane potentials of individual neurons to oscillate (Bacci et al. 1999; Misgeld et al. 1998; Murphy et al. 1992; Nunez et al. 1996; Robinson et al. 1993; Senn et al. 1998). In particular, we found that, although GABA_A receptors are activated synchronously with glutamate receptors, bath application of the GABA_A receptor antagonist picrotoxin had little effect on the periodicity of compound PSCs and [Ca²⁺]_i transients (Fig. 4). In addition, whole cell current-clamp recordings revealed no substantial afterhyperpolarizations following spontaneous depolarizations, indicating that the periodicity of the events was not determined by oscillations in the membrane potentials of ganglion cells (Fig. 3C). Rather, the periodic activity was prevented by blockers of glutamatergic synaptic transmission, and, in some cultures, cholinergic transmission, indicating that the spontaneous activity is mediated exclusively by excitatory synaptic transmission.

We presented preliminary evidence that synaptic depression may play a role in setting the periodicity of the spontaneous activity in our dissociated cultures (Senn et al. 1998; Traub et al. 1989). We found that the time course of recovery from synaptic depression closely matched the mean interval we measured between compound PSCs or [Ca²⁺]_i transients (Fig. 5). To demonstrate that the frequency of events is determined by the kinetics of recovery from synaptic depression, pharmacological manipulations that alter the kinetics of recovery without changing the efficacy of synaptic transmission are required.

Different versions of this depression mechanism have been proposed to regulate the temporal structure of spontaneous correlated activity in vivo. In the developing spinal cord, the depression that limits the length of bursts of action potentials recorded from motoneurons manifests itself as a decrease in synaptic strength (Fedirchuk et al. 1999; Tabak et al. 2000). In hippocampal networks, periodic activity is thought to be modulated by synaptic depression via postsynaptic mechanisms (Traub et al. 1989) or presynaptic mechanisms, such as the depletion of glutamate pools (Bains et al. 1999; Staley et al. 1998). In the developing mammalian retina, this form of network depression has been described as a refractory period, a finite period of time after activation of a region of the retina during which that region cannot participate in subsequent waves (Butts et al. 1999; Feller et al. 1997). In the case of the intact retina, the period of network depression is on the order of 30-40 s. Hence, if synaptic depression is responsible for this refractory period, then the mechanism would be different from the fast recovery observed in culture. Recovery from depression at synapses in hippocampal neurons can last on the order of 30 s when the readily releasable pool has been fully depleted (Liu and Tsien 1995; Stevens and Wesseling 1999). In addition, the long recovery time seen in the intact retina may be a reflection of the kinetics of depression at TTX-independent graded-release synapses (von Gersdorff and Matthews 1997) that provide the synaptic input to RGCs during waves in contrast to the TTX-dependent release observed in culture. Thus it is possible that synaptic depression sets the slow periodicity of the spontaneous activity seen in vivo.

In summary, we have demonstrated that cultured retinal neurons can act as a model system for studying the cellular mechanisms underlying spontaneous correlated periodic activity. A culture that would better represent the in vivo retinal circuit would be enriched for graded-release retinal interneurons. Whether this culture will generate activity with similar patterns remains to be determined.