Abstract
In the mammalian retina, excitatory and inhibitory circuitries enable retinal ganglion cells (RGCs) to signal the occurrence of visual features to higher brain areas. This functionality disappears in certain diseases of retinal degeneration because of the progressive loss of photoreceptors. Recent work in a mouse model of retinal degeneration (rd1) found that, although some intraretinal circuitry is preserved and RGCs maintain characteristic physiological properties, they exhibit increased and aberrant rhythmic activity. Here, extracellular recordings were made to assess the degree of aberrant activity in adult rd1 retinas and to investigate the mechanism underlying such behavior. A multi-transistor array with thousands of densely packed sensors allowed for simultaneous recordings of spiking activity in populations of RGCs and of local field potentials (LFPs). The majority of identified RGCs displayed rhythmic (7–10 Hz) but asynchronous activity. The spiking activity correlated with the LFPs, which reflect an average synchronized excitatory input to the RGCs. LFPs initiated from random positions and propagated across the retina. They disappeared when ionotrophic glutamate receptors or electrical synapses were blocked. They persisted in the presence of other pharmacological blockers, including TTX and inhibitory receptor antagonists. Our results suggest that excitation—transmitted laterally through a network of electrically coupled interneurons—leads to large-scale retinal network oscillations, reflected in the rhythmic spiking of most rd1 RGCs. This result may explain forms of photopsias reported by blind patients, while the mechanism involved should be considered in future treatment strategies targeting the disease of retinitis pigmentosa.
Introduction
Enhanced neuronal oscillations and aberrant synchrony are characteristic features of disorders of the CNS, including the epilepsies and Parkinson's disease (for review, see Uhhaas and Singer, 2006). In these diseases, an imbalance between excitatory and inhibitory circuitry may cause enhanced excitability and subsequent aberrant synchronous behavior in neuronal populations. Gap junctions—intercellular channels that directly connect neighboring cells—are thought to support such neuronal activity patterns (Rufer et al., 1996; Dudek et al., 1998; Carlen et al., 2000).
Recently, the rod-degenerated (rd1) mouse retina, a model that mimics the human disease of retinitis pigmentosa (RP), has been reported to display some electrophysiological features reminiscent of pathological oscillatory discharge (Ye and Goo, 2007; Margolis et al., 2008; Stasheff, 2008). The degeneration of photoreceptors triggers changes in the retinal morphology in these mouse retinas (for review, see Marc et al., 2003), and in human RP patients (Kolb and Gouras, 1974; Fariss et al., 2000). Neurons postsynaptic to photoreceptors (horizontal and bipolar cells) retract their dendrites and develop aberrant axonal processes (Strettoi and Pignatelli, 2000; Strettoi et al., 2002), whereas amacrine cells and retinal ganglion cells (RGCs) maintain their morphology up to late degenerate stages (Stone et al., 1992; Santos et al., 1997; Humayun et al., 1999; Strettoi et al., 2003; Mazzoni et al., 2008). On a molecular level, functional remodeling of ionotrophic and metabotropic glutamate receptors in inner retinal neurons of adult rd1 retinas has been reported (Chua et al., 2009); however, little is known about the expression patterns of inhibitory receptors or gap junctions (Fletcher, 2000).
In adult rd1 retinas, projection neurons—the RGCs—develop increased spontaneous activity that is in many cases rhythmic and maintained for weeks after light-evoked responses disappear (Ye and Goo, 2007; Margolis et al., 2008; Stasheff, 2008; Ryu et al., 2010). Indirect evidence of rhythmic retinal activity was reported earlier in superior colliculus recordings of anesthetized rd1 mice (Dräger and Hubel, 1978). It remained unclear whether the rhythmic activity occurred in a subset of RGCs, whether it is synchronized at the population level, and what are the mechanisms governing such activity.
Here, we investigate these questions by recording the electrical activity in RGC populations of adult rd1 retinas (C3H and FVB/N) using a high-density multi-transistor array (MTA) (Eversmann et al., 2003; Lambacher et al., 2004). The array comprises 16,384 electrodes packed in 1 mm2, thus allowing for very high spatial electrical imaging of the interfaced in vitro retina. Simultaneous recording from up to 100 RGCs reveals that the majority of them spike at a rhythm of 7–10 Hz. This rhythmic spiking is driven by strong presynaptic input that causes spatially extended local field potentials (LFPs). Based on the LFP propagation velocity, the LFP sensitivity to gap junction blockers, and phase-shifted RGC spiking, we postulate that electrical synapses between retinal interneurons mediate the transmission of aberrant rhythmic activity. The presented results may explain some of the neuropathological findings in RP patients and may guide strategies to treat the disease.
Materials and Methods
Animals.
All experiments were performed in accordance to the animal use committee of the Max Planck Institutes. In this study, retinas from male rd1 mice were investigated between postnatal day 35 (P35) and P70, when their retinas are no longer responsive to light. All experiments on rd1 retinas were first performed using the local colony of the FVB/NCrlMPI strain that is homozygous for the Pde6brd1 mutation (Taketo et al., 1991). After the pharmacological protocols had been established, the experiments were repeated with rd1 retinas from C3H/HeNCrl mice (Charles River) that are homozygous for the Pde6brd1 mutation. Control experiments were performed on wt retinas from the local colony of C57BL/6NMPI male mice. To rule out possible interference from different genetic backgrounds, we additionally obtained C3H wt male mice from F. Paquet-Durand (Tübingen, Germany). This mouse strain was initially created in the laboratory of S. Sanyal (Sanyal and Bal, 1973) by substituting the mutated Pde6b gene with the normal allele and has been maintained in various laboratories ever since. All animals were housed in temperature-regulated facilities on a 12 h light/dark cycle and fed ad libitum. All animals were dark adapted (1 h) before the retina preparation.
Preparation of the mouse retina and mounting on the multi-transistor sensor array.
The preparation of the retina was performed under dim red illumination (640 nm LED; Roithner Lasertechnik) that also illuminated the room during the experiment. Mice were anesthetized with isoflurane (CP Pharma) and killed by cervical dislocation. Their eyes were removed, bathed in room temperature-oxygenated Ames' medium (Sigma-Aldrich; A1420; pH 7.4), and hemisected. Next, the lens and vitreous were removed from the eyecup, and finally the retina was gently peeled off the pigment epithelium. The retina was mounted ganglion cell side down on a poly-l-lysine (150 kDa molecular weight; Sigma-Aldrich)-coated MTA. The MTA itself was glued on a ceramic package (CPGA; Spectrum). The bond wires are shielded with a custom-made Perspex chamber with an inner area of 12 mm2. During the recording, retinal tissue on the MTA was continuously perfused with oxygenated Ames' medium (33–36°C) at a rate of 7 ml/min.
Electrical recording with multi-transistor arrays.
The electrical response of the retina was measured using an array of 128 × 128 equally spaced sensor transistors covering an area of 1 mm2. For the default configuration, we measured every second column (128 × 64 sensors) with a sampling frequency of 12 kHz for each sensor. In each experiment, sensor transistors were calibrated by applying an AC voltage (frequency, 70 Hz; amplitude, 3 mV peak-to-peak) to the bath electrode (Eversmann et al., 2003; Lambacher et al., 2004). The calibration voltage changes the electrical potential at the surface of the chip. The local change of electrical potential couples through the insulating electrolyte/chip interface to the gate of the sensor transistor and proportionally modulates the source-drain current therein. During the experiment, ion currents through excited retinal ganglion membranes change the local extracellular voltage with respect to the bath electrode. The potential at the chip surface couples through the insulating electrolyte/chip interface to the gate and proportionally modulates the source-drain current. The response of each sensor transistor is solely determined by the potential above the insulating TiZrO2 layer, averaged over the diameter (6.3 μm) of the top contact. The insulating TiZrO2 layer had a thickness of ∼30 nm. The chip readout pattern was optimized to avoid cross talk of transistor signals on the chip (Eversmann et al., 2003; Lambacher et al., 2004). During the recording, the columns of the sensor array were sequentially connected to 128 line amplifiers. After a settling time of 720 ns, the output of these line amplifiers was multiplexed over another 640 ns into 16-output channels. The readout time of 128 × 64 sensor array was therefore approximately (1.36 × 64) μs. Within each sensor column, an 8:1 multiplexer selects 16 sensors (sensor spacing, 125 μm) that are read out within ∼640/8 ns.
Identification of action potentials and assignment to the corresponding ganglion cell.
The method for identifying action potentials and assignment to corresponding neurons has been described in a recent report (Lambacher et al., 2010). Briefly, the analysis is done in three steps: (1) identification of threshold crossings of a signal vector V calculated from neighboring extracellular voltages, (2) assignment of threshold crossings to one action potential, and (3) assignment of action potentials to corresponding neurons.
For the identification of threshold crossings, we first apply a bandpass filter to the calibrated data (0.1–3 kHz). As the sensor distance (7.8 μm) is smaller than the ganglion cell soma size, each extracellular signal is picked up by more than one sensor. The duration of the somatic extracellular signal is longer than the time interval between consecutive data points. Therefore, for each recorded data point, the length of a signal vector V is calculated as follows:
with Vi representing the signal amplitude of data point i in neighborhood; σi, root mean square (rms) noise of transistor in neighborhood. The sum runs over a 3 × 3 × 3 neighborhood (three sensor rows, three sensor columns, three time points) surrounding the data point under consideration. The data point itself is part of the neighborhood. If V exceeds a threshold of 15, the data point is saved and considered part of the extracellular waveform that represents the action potential. Assuming equal noise on each of the nine neighboring sensors and homogenous coupling on these sensors, the threshold value of 15 means that those extracellular voltages exceeding 15/√27 × rms of the corresponding sensor are detected. This threshold value is close to that of previous studies using metal electrode arrays (Zeck and Masland, 2007; Stasheff, 2008) but slightly higher than the value of 11.7/√27 × rms used in the study by Lambacher et al. (2010).
In a second step, threshold crossings are combined to action potentials. All threshold crossings that are spatially adjacent at the same time point are merged into a “cluster.” Next, we consider the spatial overlap of time-consecutive clusters. Two such clusters are part of the same action potential if they share at least one sensor. All clusters that belong to an action potential are combined and the data point (time stamp and sensor location) with the highest amplitude is chosen as a representative for the action potential.
Finally, action potentials are assigned to the corresponding ganglion cells. Action potentials recorded on one sensor may belong to different cells. We again take advantage of the high spatial sampling and align the centers of gravity of two action potentials in time. For each action potential we consider the extracellular voltages surrounding the center of gravity. The cross-correlation (CC) between the neighborhoods of action potentials i and j is calculated as follows:
with
as the normalization factor for action potential i. The index k ranges over all peaks that are common to both action potentials i and j, whereas the index l ranges over all voltages that constitute the action potential. The following analysis separates action potentials that originate from different neurons. For a number of M overlapping action potentials, a symmetric matrix of cross-correlation values cij(i,j = 1:M) is obtained. This matrix is rearranged to minimize the cij differences between adjacent rows. If all action potentials under consideration belong to one neuron, we obtain little variation within the correlation matrix. If action potentials from several neurons are compared, separate clusters are visible in the cross-correlation matrix. This matrix is then split and the action potentials within each cluster are assigned to a different neuron. The sorting of action potentials is comparable with the supervised algorithm used in a previous study, although, there, a different algorithm (k-means clustering) was used (Zeck and Masland, 2007). The spike sorting and splitting is done off-line and semiautomated. The final spike trains are tested to obey a refractory period of at least 1 ms. No action potentials with interspike intervals <1 ms were assigned to one neuron using the described method.
Spike train analysis.
Data were analyzed using custom software written in Matlab (The MathWorks). The average firing rate for each spontaneously active RGC was computed as the total number of spikes divided by the length of the recording period. The recording consists of 1–10 s concatenated segments of continuous voltage traces (0.1–12 kHz sampling frequency). The fundamental spiking frequencies were estimated from the autocorrelation functions.
CC functions were computed for cell pairs after spike trains had been assigned to particular cell. Normalized CCs were calculated using the Matlab routine xcorr. This routine calculates the dot product of two normalized vectors representing the RGCs spike trains. The spike trains were binned with either 4 or 0.4 ms time resolution. The correlation coefficient (Pearson's correlation) presented here represents the zeroth lag of the correlation function. It represents the percentage of spikes from one spike train that occurs in the same time bin in the correlated spike train and therefore represents an estimate of the coupling strength between two RGCs. The length of correlated spike trains (∼5 min in 4 ms time bins) sets the statistical significance of the correlation coefficient <0.01.
Local field potentials and propagation velocity.
Extracellular voltage changes characterized by negative deflections (∼20 ms long) followed by slower repolarization (∼100 ms) have been reported as a slow-wave component in a recent rd1 study using multielectrode arrays with large electrode distances (Ye and Goo, 2007; Ryu et al., 2010). Voltage modulations that occur in phase across neighboring sensors—but not across the whole sensor array (1 mm2)—reflect spatially confined local field potentials. Negative deflections in the extracellular potential are caused by the depletion of positive ions or by the accumulation of negative ions in the extracellular space and reflect membrane depolarizations of neurons in the ganglion cell layer.
The LFP fundamental frequency was estimated from the power spectral density (PSD) functions computed from long (10 s) calibrated voltage traces. The fundamental frequency was measured as the first peak power in the range 3–100 Hz.
The calculation of LFP propagation velocity is performed in analogy to the velocity calculation of developmental calcium waves (Blankenship et al., 2009). The method requires accurate definition of the LFP boundary. Therefore, each voltage map (1 mm2) recorded by the sensor array at one time point was first spatially filtered (Gaussian filter, σ = 10 μm). Time-consecutive voltage maps were averaged within a window of 2 ms. In each such averaged voltage map, we identified those sensor areas that measured voltages smaller than −100 μV as a “region of interest” (ROI). Contiguous regions were considered to be part of one LFP. The next ROI was calculated after a delay of 10 ms. LFP propagation velocity was computed if consecutive ROIs covered at least one-half of the sensor array (1 mm2) and lasted >50 ms. The border of each ROI represents an isotemporal continuous line of the LFP wavefront. The furthest distance traveled during the last time interval was used as an ending point of the propagation path. Points along the propagation path were selected by finding the shortest distance between the point at time t (in milliseconds) and the isotemporal line at time t − Δt. Velocity was calculated by averaging the distance between consecutive time points Δt (values of Δt are specified in the legends of Figs. 3, 5, 6). To clarify that the velocity values did not depend on the threshold (−100 μV) or the time steps Δt, we calculated velocities for a subset of waves using a threshold of −50 μV and time steps between 7 and 15 ms, respectively. We obtained qualitative similar results.
Statistical analysis.
We could not rigorously test whether the distributions of the various parameters (firing rate; correlation coefficients; LFP velocity; maxima of the power spectral density) follow a normal distribution. To test for statistical significance, we therefore compared median values using the Wilcoxon–Mann–Whitney U test. For the firing rate and correlation coefficient, we present mean values as they are not different from the median.
Pharmacology.
To block AMPA/kainate receptors, 6,7-dinitroquinoxaline-2,3-dione disodium salt (DNQX) was used. NMDA receptors were blocked by dl-2-amino-7-phosphonoheptanoic acid (AP-7). Glycinergic synapses were blocked by strychnine, whereas for GABAA receptors SR-95531 hydrobromide (gabazine) [6-imino-3-(4-methoxyphenyl)-1(6H)-pyridazinebutanoic acid hydrobromide] was used. Gap junctions were blocked by either meclofenamic acid (MFA) or carbenoxolone (CBX). All antagonists were purchased from Tocris Bioscience. Sodium channels were blocked using tetrodotoxin (TTX) (Sigma-Aldrich).
Results
We recorded spontaneous electrical activity from isolated adult rd1 retinas using a multi-transistor sensor array (Eversmann et al., 2003; Lambacher et al., 2004). The sensor distance of 7.8 μm represents a considerable improvement over existing multielectrode arrays and allows for the precise spatial mapping of electrical activity in the retinal ganglion cell layer. Extracellular signals originating from RGC action potentials were identified using an appropriate filter range (0.1–3 kHz) and taking advantage of the simultaneous recording on many adjacent electrodes. In the low frequency range (1–60 Hz), we detected extracellular voltage modulations that are caused by the depletion and accumulation of ions in the extracellular space, respectively. Low-frequency changes of the extracellular potential that appear simultaneously on adjacent sensors constitute local field potentials.
The majority of identified retinal ganglion cells in rd1 retinas exhibit rhythmic activity
Mouse models of retinal degeneration mimic the disease of retinitis pigmentosa, both on the anatomical and physiological level (Chang et al., 2002). Although the gross morphology of the retinal output neurons—the ganglion cells—remains intact in the rd1 mouse (Strettoi et al., 2002; Margolis et al., 2008; Mazzoni et al., 2008), recent studies report increased spontaneous activity in rd1 ganglion cells (Ye and Goo, 2007; Margolis et al., 2008; Stasheff, 2008) and rhythmic spiking activity in three morphology identified cell types (Margolis et al., 2008). Here, we confirm and extend this finding in adult rd1 retinas (P35–P70) from two different rd1 strains (C3H and FVB/N, respectively).
The spontaneous activity of 20 selected RGCs recorded simultaneously in one retinal portion is shown in Figure 1a. The presented RGCs display a rhythmic bursting activity visible in the spike trains autocorrelation function (Fig. 1b). High-pass-filtered (0.1–3 kHz) calibrated voltage traces of selected RGCs are shown in supplemental Figure 1 (available at www.jneurosci.org as supplemental material). Some of the rhythmic RGCs do not elicit bursts continuously; however, their autocorrelation functions remain rhythmic. Within the retina presented in Figure 1a–d, the spiking of most RGCs (90%) occurred with the same fundamental frequency (7.5 ± 0.3 Hz; mean ± SD; n = 70 RGCs). The average fundamental frequency calculated for 403 rhythmic RGCs (n = 7 retinas) was 9.2 ± 1.8 Hz (mean ± SD). This result was obtained from both rd1 mouse strains used in this study. RGCs in older retinas (C3H/HeNCrl; P160–P200; n = 110 RGCs in three retinas) display rhythmic activity with an average fundamental frequency that is not significantly different from the values calculated for the younger retinas (8.8 ± 1.3 Hz; mean ± SD; p = 0.22). The percentage of rhythmic RGCs is high across all rd1 retinas (85% of RGCs in P35–P70) and declines slightly in older retinas (75% of RGCs in P160–P200). The average firing rate (Table 1) of spontaneous activity in rd1 RGCs (P35–P70) is significantly higher (p < 0.01) than in old rd1 RGCs (P160–P200), in agreement with previous reports (Margolis et al., 2008; Stasheff, 2008). The absolute value of spontaneous firing rate in rd1 RGCs is in line with Margolis et al. (2008) but higher than the value reported by Stasheff (2008). This may be attributed to slightly different recording conditions such as perfusion medium or perfusion rate.
The majority of RGCs in the rd1 mouse retina exhibit rhythmic spiking. a–d, Spike train properties of RGCs recorded in a 1 mm2 portion of an rd1 retina. a, Spiking activity from 20 selected RGCs. Each tick represent the occurrence of one action potential. The three RGC spike trains in the top row are further evaluated. b, Autocorrelation functions of the three selected rd1 RGCs reveal rhythmic activity with an average interval of ∼150 ms. The peak at zero time lag is omitted. c, The spike train CCs between three selected RGCs reveal strong oscillations. For two cell pairs, the activity is phase-shifted. One RGC pair fires in synchrony revealed by the central peak at zero time lag. Bin size, 4 ms. d, The same CCs as shown in c, computed at higher resolution (bin size, 0.4 ms). A double peak in one CC at zero time lag indicates electrical coupling between the two RGCs. e–h, RGC spike train properties in wt retinas. e, Raster plot of spontaneous activity. f, Autocorrelation function of three selected RGCs. g, Spike train cross-correlations between the three selected RGCs shown in f. Synchronous activity is detected in one cell pair. Bin size, 4 ms. h, The same CCs shown in g, computed at higher resolution, reveal a double peak in one CC around origin, similar to the result in the rd1 retina (d). Bin size, 0.4 ms.
Properties of retinal ganglion cells and local field potentials in rd1 retinas
We could not detect any rhythmic activity in spontaneously active RGCs from wt retinas (Fig. 1e,f) recorded with the same multi-transistor array under otherwise identical recording conditions (n = 200 RGCs; n = 7 retinas; age, P35–P60). This result applied to both wt mouse strains used in this study. We note, in agreement with previous studies (Margolis et al., 2008; Stasheff, 2008), that the spontaneous firing rate of wt RGCs is significantly lower (p < 0.01) than the firing rate of rd1 RGCs of the same age (Table 1).
The majority of rhythmic rd1 RGCs display phase-shifted activity
In the following, we investigate whether all RGCs in one retinal portion fire in synchrony. The cross-correlograms (CCs) between pairs of rd1 RGC spike trains display multiple peaks (Fig. 1c) separated by 100–150 ms. In most CCs (86%; n = 1688 analyzed CCs within one retina), the correlation peak does not occur at zero time lag (Fig. 2) but at any value between −70 and 70 ms. This variability suggests that either the RGCs in the recorded retinal portion do not receive the same synchronous input or that each RGC exhibits a different spike threshold.
Nearby RGCs in rd1 retinas oscillate with little time lag. a, Histogram of the time lags of the central CC peaks of RGC spike trains calculated within one rd1 retina (2 CCs were presented in Fig. 1c). The dashed line marks chance level. The dark gray bars mark the result obtained if the first half of recording is considered; the light gray bars, the results obtained from the second half of recorded data. Bin size, 10 ms. Each CC time lag is normalized to the average rhythm period. b, Average of five CC histograms calculated as in a. With the exception of zero time lag, there is no preference for any phase shift of the central CC peak. Each gray line represents the result from one retina. The thick gray line represents the data shown in a. The black symbols mark the mean values for a given time lag. The dashed line represents uniform distribution. c, Dependence of the peak CC time lag on the distance between RGCs. The analysis was performed on the same dataset shown in a. Up to separation distance of ∼300 μm, the median time lag increases with increasing distance. The black bars denote the median values obtained for the whole recording session. The dark gray bars mark the result obtained if the first half of recording is considered; the light gray bars, the results obtained from the second half of recorded data. No significant changes are measured. Bin size, 50 μm. d, Average of five histograms of median time lags calculated as in c. The same dataset of the five retinas evaluated in b was used. The thick gray line represents the data shown in c. The black symbols mark the median time lag for a given RGC distance. The tendency of short time lags for RGC separations <500 μm is preserved across retinas.
We first analyze CCs that show a central peak at zero time lag. When computed at high resolution, several “peaked” CCs displayed two subpeaks around zero time lag (Fig. 1d). The two maxima were located on average at −1.4 ± 0.2 and 1.3 ± 0.3 ms, respectively (n = 40 pairs in three retinas). These peaks are a strong indicator of reciprocal electrical coupling between RGCs, where the spike in one cell depolarizes the coupled cell above spiking threshold. Comparable double-peaked synchronization patterns have been described in electrically coupled RGCs in healthy mammalian retinas (Mastronarde, 1989; Hu and Bloomfield, 2003). The average distance between electrically coupled rd1 RGCs is 120 ± 30 μm (range, 60–180 μm). This indicates that only nearby RGCs are electrically coupled. The average correlation strength of these pairs is 0.20 ± 0.05 (mean ± SD; bin width, 4 ms). The correlation strength does not vary over the distance mentioned above. In CCs calculated from wt RGC spike trains, we also find peaked CCs (n = 15) that display two subpeaks around zero time lag (Fig. 1g,h). The CC maxima were located on average at −1.8 ± 0.4 and 1.7 ± 0.3 ms, respectively, whereas the spacing between cells was 133 ± 40 μm (range, 66–224 μm). The average correlation strength of these pairs is 0.10 ± 0.04 (mean ± SD; bin width, 4 ms). The correlation coefficients in wt RGCs are statistically different (p < 0.001) from those measured for rd1 RGCs. We conclude that the subpopulation of electrically coupled RGCs in rd1 and wt retinas have similar properties, except for the stronger electrical coupling strength between rd1 RGCs.
We return to the analysis of the rd1 spike train CCs. As mentioned, the majority of spike train CCs (range, 60–86%; n = 5 retinas) have their most central peak shifted with respect to zero time lag (Figs. 1c, 2). We therefore tested whether there was any preferred phase shift between RGC spiking and whether the phase shift depends on the RGC separation. As the fundamental RGC spiking frequency differs among retinas (range, 7–10 Hz), we calculate for each retina an average rhythm. We then normalize each CC time lag to the retina specific rhythm. A time lag of 50 ms between two RGCs spiking at a fundamental frequency of 10 Hz corresponds to a relative shift of period/2, whereas the same absolute time shift gives a smaller relative shift for RGCs spiking at 7 Hz. First, the probability to measure a shifted CC peak is close to chance level for any value within one period, with the exception of zero time lag (Fig. 2a,b) (n = 5 retinas). The average probability varies little if either half of the recording session is evaluated (Fig. 2a), although for individual CCs a small shift of the central peak is observed (data not shown). Second, for all RGCs separated by <500 μm, the median value of the central CC peaks increases from 8 ms (RGC separation, <50 μm) to 25 ms (RGC separation, 500 μm) (Fig. 2c). For separation distances >500 μm, the peak time lags occur at any value with nearly equal probability (Fig. 2d).
These findings point toward a local driving force of the rhythmic spiking of rd1 RGCs. In the next paragraph, we therefore investigate whether this driving force is caused by independent, “pacemaker-like” cells, as might be inferred from Figure 2, a and b, or whether spatial interactions (suggested by Fig. 2c,d) shape the oscillatory behavior.
The rhythmic spiking of rd1 RGCs correlates with local field potential minima
All sensors recording rhythmic rd1 RGC spiking display a modulation of the extracellular voltage in the low-frequency (1–60 Hz) range (Fig. 3a). Voltage modulations that occur in phase across nearby sensors reflect spatially confined local field potentials (Fig. 3b). Negative deflections in the extracellular potential are caused by the depletion of positive ions or the accumulation of negative ions in the extracellular space, as reported in other neural tissues (Mitzdorf, 1985). LFPs were never found in wt retinas under the same recording conditions.
RGC spiking and local field potential minima coincide in rd1 retinas. a, Extracellular voltage recordings from two sensor electrodes separated by ∼300 μm. The filter settings of 1 Hz to 3 kHz reveal single spikes and a slow oscillatory extracellular potential. The sensor locations are marked in subplot b1. Red trace, Low-pass-filtered (1–60 Hz) signals reveal rhythmic LFPs. The horizontal arrow marks zero extracellular potential. The open green arrows mark the start point and end point of a six frame series of extracellular voltage maps, shown in b. b, Extracellular voltage maps (“electrical images”) recorded at a spatial resolution of 8 × 16 μm. Each image represents the average extracellular voltage over 2 ms. Separation between images is 10 ms. Scale bar, 200 μm. c, Cross-correlation of three spike trains with LFP minima revealing central peaks with minimal time lag (gray). The average of the three traces is shown as a thick black line. LFP minima were evaluated for sensors separated by 32 μm from the RGC recoding sensor. d, Distribution of peak CC time lags for RGCs and nearby recorded LFPs in two retinas. The recordings presented here were from a different rd1 retina as shown in Figure 1.
LFPs initiated at several locations across the rd1 retina. During the recording time (1–3 h), the entire portion of the imaged ganglion cell layer was part of a LFP. Continuous (10 s) readout of the sensors was used for PSD analysis indicating a fundamental LFP frequency of 7–10 Hz (Table 1). LFP and spiking activity was interleaved with periods of silence (supplemental Fig. 1, available at www.jneurosci.org as supplemental material).
To investigate whether LFPs are responsible for rhythmic RGC spiking, we correlated the spike trains of individual RGCs with the timing of LFP minima recorded in the vicinity of the ganglion cells. Different temporal filters for action potential (0.1–3 kHz) and LFP (0–60 Hz) identification, and a distance of 32 μm between the RGC recording sensor and the LFP recording sensor guarantees that RGC spikes do not influence the LFP minima. The cross-correlation of three selected LFP–RGC pairs revealed an oscillatory function peaked around zero lag (Fig. 3c). The analysis of 188 LFP–RGC pairings in three retinas reflects that most CC peaks occur in the interval around zero lag. The majority (75%) of the CC peaks occur at a time lag between −12 and 2 ms (Fig. 3d).
Propagation of local field potentials
Local field potentials are dynamic: they expand (Fig. 4) or propagate across the retinal surface (Fig. 3) but never collapse. Occasionally, we observed the initiation of two LFPs within the imaged area. Within one retina, consecutive wave-like LFPs propagated in the same direction (Fig. 3b). Because the propagation of LFPs resembled activity patterns seen in immature retinas, we adopted methodology used to assay the speed of developmental waves (Blankenship et al., 2009). We consider LFPs averaged over 2 ms of electrical activity. In each such averaged voltage map, we identified those sensor areas that measured voltages smaller than −100 μV as a ROI. The next ROI was calculated after a fixed delay Δt (Fig. 4). The identified ROIs propagated across the retina with a median velocity of 8 mm/s (n = 292 waves in five retinas; for details of velocity calculation, see Materials and Methods). The propagation of electrical activity could, in principle, be assessed using the RGC spiking. However, because RGCs started a burst (3–10 action potentials) every 100–150 ms, the methodology used in extracellular recordings of developing retinas (Demas et al., 2003), which display much longer interburst intervals, failed here.
Local field potentials propagate across the retina. a, Extracellular voltage maps recorded at time intervals of 15 ms. Each image represents the average activity over 2 ms. The activity was recorded in the same retina shown in Figure 3b, but recorded at a later time. b, For the estimation of the wave-like LFP propagation velocities (see Materials and Methods), we identified region of interest with voltages more negative than −100 μV. Each binary image represents the region of interest calculated for each voltage map shown in a. Right frame, Spatial extension of LFP propagation. Each grayscale represents the region of interest at one of the five time points. Two arbitrary points are shown that were selected to trace back the propagation paths (red). c, Histogram of LFP propagation velocities. A total of 292 LFPs within five retinas were evaluated. Bin size, 2.5 mm/s. Median value, 8 mm/s.
The analysis of RGC spiking and LFP propagation suggests that RGC depolarizations occur through a local mechanism that spreads laterally across the retina. To identify circuitry involved in the generation and propagation of LFPs and concomitant RGC spiking, we applied pharmacological agents that disrupted either RGC spiking and/or LFPs.
Local field potentials persist when voltage-gated sodium channels or inhibitory receptors are blocked
We first excluded the hypothesis that spontaneous sodium spikes are responsible for the extracellular LFPs by adding the sodium channel blocker TTX (0.2 μm) to the perfusion solution. Under these conditions, we measured LFPs but no RGC spikes (Fig. 5a,b). The sodium channel blocker abolished the RGC firing in all but seven RGCs from a total of 138 identified cells in three retinas. The RGC firing rate dropped to 0.3 Hz (n = 7 RGCs in three retinas). The LFP amplitude measured as the peak of the voltage power spectrum (Fig. 5b) did not change after application of TTX. We quantified LFP amplitudes on 20 sensors that recorded from the same retina before and after the application of TTX. The average of the maximum power spectral densities was not different (p = 0.1) in the two conditions (Fig. 5b, inset). Similarly, no visible changes occur in the overall LFP shape (Fig. 5c). However, the LFP propagation velocity (Table 1) decreased to 6.6 ± 2.6 mm/s (mean ± SD; n = 55 waves; statistical significance, p < 0.05) (Materials and Methods). The LFP fundamental frequency decreased from an average of 9 to 6.2 Hz calculated on 16 sensors in the three retinas (one trace shown in Fig. 5b). After washout, the rhythmic sodium spikes appeared again in all three retinas tested. This experiment proves that, although sodium channels shape certain LFP properties, the RGC action potentials do not contribute to LFPs. Thus, neurotransmitters acting on nonspecific ion channels may be responsible for the LFP generation.
Application of the sodium channel blocker TTX abolishes RGC spikes but does not inhibit the initiation and propagation of LFPs. a, Extracellular voltage recording on one sensor, showing the RGC spikes and LFPs under control conditions (top trace) and after the application of 0.2 μm TTX (red trace; bottom). b, Power spectral density of the traces shown in a demonstrate that TTX reduces the fundamental LFP frequency. The amplitude of the LFPs, reflected in the PSD, does not change significantly after the application of TTX. Inset, Average peak power measured on 30 sensors before (black) and after (red) the application of TTX. c, Representative electrical images of TTX treated rd1 retina indicating the LFP expansion. Each frame represents the electrical activity averaged over 2 ms. The time interval between frames is 15 ms. The traces shown in a were selected from a sensor in the center of the array.
In a second experiment, we investigated the contribution of inhibitory circuits on the generation of LFPs and rhythmic RGC activity. In the rd1 retina, the two inhibitory cell classes are preserved to a very different degree—horizontal cells retract axonal and dendritic processes, whereas amacrine cells appear morphologically well preserved (Strettoi and Pignatelli, 2000; Park et al., 2001; Strettoi et al., 2002). Amacrine cells inhibit synaptically connected bipolar cells or RGCs by the release of either GABA or glycine (Wässle, 2004). Horizontal cells modulate photoreceptor output through feedback mechanisms that are still debated (Wässle, 2004). The application of the combination of glycinergic receptor blocker strychnine (20 μm) and GABAA receptor blocker gabazine (40 μm) had the following effects on RGC spiking and LFPs. The average firing rate (Table 1) of RGCs did not change significantly (p = 0.09) in the presence of inhibitory receptor blockers. Blocking inhibitory receptors led to the synchronization of more RGCs (Fig. 6b,c). The central peak in the spike CCs occurred with a time lag of <5 ms in 43% of the CCs (average over RGCs in five retinas) compared with 23% in untreated retinas (Fig. 2b). Within each retina, the probability of synchronous RGC activity increased on average twofold. RGC spike trains and LFP minima correlated with low time lag in the presence of inhibitory receptor blockers, as was observed for the unperturbed state (compare Figs. 6d,e, 3c,d). The spatial extent and extracellular amplitude of LFP increased after disinhibition of the retina (Fig. 6f). However, we never measured simultaneous low-frequency oscillations covering the whole sensor array (1 mm2), both in disinhibited and also in unperturbed rd1 retinas. The fundamental RGC spiking frequency (Table 1), the LFP frequency, and the LFP conduction velocities decreased compared with the unperturbed state significantly (6 ± 2 mm/s; mean ± SD; n = 110 waves; p < 0.05). The application of each inhibitory blocker alone revealed that strychnine (2 μm; n = 2 retinas) had a stronger effect on the LFPs than gabazine (40 μm; n = 2 retinas) (supplemental Fig. 2, available at www.jneurosci.org as supplemental material). The average LFP amplitude—measured as the peak of the voltage power spectrum—increased significantly (p < 0.001) when glycinergic receptors were blocked. The LFP amplitude did not increase significantly when only GABAA receptors were blocked (p = 0.07). Strychnine reduced the fundamental LFP frequency to a larger extent than gabazine (supplemental Fig. 2, available at www.jneurosci.org as supplemental material).
Inhibition of glycinergic and GABAergic receptors increases the LFPs and the synchronization among rd1 RGCs. a, Top traces, Spike trains from six RGCs recorded after the application of inhibitory receptor blockers. Each tick marks the occurrence of one spike. The sensor locations are marked in subplot f. Bottom traces, LFPs measured near the spike recording sensors (bandpass of 1–60 Hz) at three positions (indicated by numbers) on the array. b, An example spike train cross-correlogram (CC) before (orange) and after the application of inhibitory receptor blockers (black trace). The CC peak shifts to zero lag when inhibitory receptors are blocked. The CCs were calculated for the spike trains of cells 3 and 4. c, Average probability distribution of minimum time lag of the central CC peak under control conditions (orange) (Fig. 2b), and with inhibitory receptors blocked (black). The distributions are calculated for the same five retinas (gray) that were evaluated in the untreated retinas (Fig. 2b). The uniform distribution is shown as a dashed line. d, CC calculated between LFP minima and RGC spike trains for the three LFPs and RGCs shown in a. e, Distribution of peak time lags for CCs calculated between RGC spike trains and nearby LFPs before (orange) and after (black) the application of glycinergic and GABAergic receptor blockers. The probability distribution was calculated for RGC–LFP pairs in one retina. f, Propagation of the LFPs in the presence of inhibitory receptor blockers. Each frame shows the electrical image averaged over 2 ms (see Materials and Methods). Scale bar, 200 μm.
These results indicate that, although inhibitory neurons shape the physical LFP properties, they do not contribute to their generation. The results suggest that, in the unperturbed rd1 retinas, glycinergic and GABAergic inhibition prevents the recruitment of additional excitatory neurons that depolarize the RGCs.
Local field potentials require glutamatergic input to RGCs
The excitatory neurotransmitter glutamate is used at several types of synapses in the retina: in the outer retina at the photoreceptor–bipolar cell synapse and in the inner retina at the synapses between bipolar cells, amacrine cells, and ganglion cells (Thoreson and Witkovsky, 1999; Wässle, 2004). In the retina, glutamate is sensed by ionotrophic (AMPA/kainate and NMDA) receptors and by metabotropic receptors. We therefore checked whether glutamate induced transmembrane currents are responsible for rhythmic RGC spiking. The application of a ionotrophic glutamate receptor (iGluR) blocker mixture (100 μm DNQX/20 μm AP-7) led to the disappearance of LFPs (compare Fig. 7a,c) (n = 4 retinas). This change was accompanied by a significant decrease of the RGC firing rate (Table 1) and the disappearance of rhythmic spike trains. After the washout of the receptor blockers, the majority of RGCs (80% in the considered RGC population) recovered their rhythmic activity. In control wt retinas, the application of the iGlu receptor blockers abolished the RGC spontaneous activity.
Inhibition of ionotrophic glutamate receptors abolishes LFPs and the rhythmic RGC spiking. a, Recording of two rhythmic RGCs and low-frequency extracellular voltage changes (bandpass of 1–60 Hz) in the untreated rd1 retina. The RGC spikes (ticks in second and third row) occur at the minimum of the extracellular voltages recorded on nearby sensors (first and forth rows, respectively). The RGCs are part of the recording shown in Figure 1a–d. b, CCs of the two rd1 RGC spike trains (shown in a) reveal the rhythmic activity centered near zero time lag. Left panel, CC calculated with a bin width of 4 ms. Right panel, CC calculated with a bin width of 0.4 ms. c, The LFPs and RGC spike trains at the same positions as shown in a after the application of ionotrophic glutamate receptor blockers (DNQX and AP-7). The low-frequency voltage modulation disappears. The RGC spiking is not rhythmic any more and occurs at a lower rate. d, CCs of the RGC spike trains shown in c reveal synchronous activity centered at zero time lag. Left panel, CC calculated with a bin width of 4 ms. Right panel, CC calculated with a bin width of 0.4 ms reveals a double-peaked CC characteristic of electrical coupling.
Why do rd1 RGCs continue to spike after inhibition of their major presynaptic input? The retinas treated with the iGluR blocker mixture revealed electrically coupled RGCs that are not seen in the unperturbed rd1 retina. The strong presynaptic input common to neighboring RGCs may obscure the electrical coupling in rd1 retinas (Fig. 7b). Inhibition of the glutamatergic input reveals a sharp peak in the CC at zero time lag (Fig. 7d), whereas in the untreated retina the same CC displayed a broad peak (Fig. 7b). When computed at high resolution, the peaked CC displayed two subpeaks around zero lag similar to those seen in unperturbed rd1 and wt retinas (Fig. 1d,h). The maxima for this pair and those of 39 others in three retinas were located on average at −1.5 ± 0.3 and 1.4 ± 0.3 ms, respectively. The average distance between electrically coupled rd1 RGCs in the presence of iGluR blockers is 128 ± 38 μm. The average strength of 0.22 ± 0.06 (bin size, 4 ms) is similar to the strength without iGluR blockers (p = 0.26). We conclude that rd1 RGCs exhibit nonrhythmic spiking in the presence of iGluR blockers that may be attributed to stronger electrical coupling between nearby RGCs (and possibly retinal interneurons) compared with wt RGCs (coupling strength, 0.01). We infer here the increased coupling strength from higher cross-correlation strength.
The application of AMPA/kainate receptor blockers alone (DNQX) had qualitatively similar results as those reported above. In contrast, the application of the NMDA receptor blocker AP-7 (100 μm) alone did not lead to any significant changes in either RGC spiking or LFP properties (n = 2 retinas), indicating that AMPA/kainate receptors are sufficient to mediate the oscillatory behavior. Metabotropic glutamate receptors of type 6 (mGluR6) are expressed in the wt retina at the sign-inverting synapse between photoreceptors and ON bipolar cells (Wässle, 2004). Immunohistochemical evidence suggest a reduced expression of this receptor in rd1 retinas (Chua et al., 2009), whereas a recent functional study using resensitized cones in rd1 retinas proves that the ON pathway is still functional (Busskamp et al., 2010). The mGluR6 receptor antagonist AP-4 (50–100 μm) had no effect on either the spontaneous RGC activity (n = 2 retinas) or on LFPs (data not shown). In summary, this set of experiments indicates that AMPA/kainate receptors on the RGCs elicit membrane currents that are reflected in our recordings as LFPs. In the following section, we search for a mechanism mediating the LFP propagation.
Gap junctional coupling is required for the propagation of local field potentials
The median LFP propagation speed of ∼8 mm/s is higher than the activity propagation measured for developmental retinal waves. Given that the fastest (stage I) developmental retinal waves are inhibited by gap junction blockers (Syed et al., 2004), we tested whether LFP propagation measured here also relies on gap junction coupling.
We investigated the effect of two gap junction blockers, CBX and MFA, on the RGC spiking activity and the occurrence of LFPs. Application of CBX (n = 3 rd1 retinas) at a concentration of 100 μm abolished the LFPs and the rhythmic RGC spiking in rd1 retinas (Fig. 8a). In the presence of CBX, 50% of the ganglion cells (82 of 162) maintained an average firing rate of 7 Hz; however, the spiking was no longer rhythmic. Washout of CBX (20–30 min) restored the LFPs and the rhythmic RGC spiking activity (Fig. 8a, right panel). However, not all RGCs recorded before CBX application were identified after the washout. We therefore repeated experiments using the new gap junction blocker MFA (100 μm) (Veruki and Hartveit, 2009). The application of MFA (n = 3 rd1 retinas) completely abolished the LFPs and the RGC spiking in rd1 retinas (Fig. 8b). Washout of MFA (∼15 min) restored the LFPs and the rhythmic RGC spiking activity (Fig. 8b). All RGCs recorded before the application of MFA were again identified after the washout of MFA. The percentages of rhythmic RGCs before and after the washout of the two gap junction blockers are summarized in Figure 8c. To check whether RGCs are capable of spiking in the presence of MFA, we added glutamate (300 μm; n = 2 rd1 retinas) to the perfusion solution. After the application of glutamate, a large number of RGCs (60 of 90) started to elicit action potentials at an average firing rate of 5 Hz. However, bath-applied glutamate did not restore LFPs and did not elicit rhythmic RGC spiking in the presence of MFA (Fig. 8d). The spike train autocorrelograms and pairwise CCs did not display any peaks (data not shown), indicating that RGCs spike independent. The application of CBX or MFA (both at 100 μm) to wt retinas did not change the spontaneous RGC firing rate. We conclude that glutamate release in the presence of functional electrical synapses is required for the aberrant LFPs and concomitant rhythmic RGC spiking.
Application of gap junction blockers abolishes LFPs and the rhythmic RGC spiking. a, Effect of the gap junction blocker CBX on RGC spiking and LFPs. Left panel, Raster plot of three rhythmic rd1 RGCs and the LFPs measured 32 μm from the soma recordings. The numbers on the left indicate the correspondence between RGCs and LFPs. Middle panels, The application of 100 μm CBX abolished rhythmic spiking and reduced the firing rate in the RGCs. The panels represent the recording 3 min (left) and 6 min (right) after the application of the CBX. Rhythmic LFPs disappeared gradually during the wash-in of the drug. Right panel, Spike raster plot from the rd1 RGC spikes shown in the left panel measured 20 min after washout of CBX. Rhythmic RGC spiking and LFPs reappear. b, The gap junction blocker MFA has similar effects as CBX. Left panel, Control recording of three rd1 RGCs and nearby LFPs. Middle panels, The application of 100 μm MFA abolished rhythmic spiking and reduced the firing rate in the RGCs. The panels represent recordings 3 and 6 min after the application of MFA. Rhythmic LFPs disappeared gradually during the wash-in of the drug. Right panel, Rhythmic RGC spiking and LFPs reappear after the washout of the drug. c, In the presence of either CBX or MFA (gray bar), the percentage of rhythmic cells dropped from ∼80% to zero but recovered after washout of the drug. After washout of CBX (n = 3 retinas), 60% of RGCs were rhythmic. After washout of MFA (n = 4 retinas), 64% of the RGCs were rhythmic. d, Left panel, Application of glutamate (300 μm) in the presence of MFA (100 μm) recovered nonrhythmic spiking in rd1 RGCs but not the LFPs. The experiment was performed on the same retina shown in b. Right panel, Raster plot of the same three rhythmic rd1 RGCs shown in the left panels and nearby LFPs measured ∼30 min after washout of MFA and glutamate.
Discussion
The major finding of our study is that, in rod-degenerated, blind mouse retinas, the majority of RGCs exhibit rhythmic spiking that is driven by spatially extended wave-like presynaptic input. The occurrence and propagation of synchronous RGC membrane depolarizations, reflected as LFPs, relies on electrical coupling between retinal neurons.
Phenomenology of the rhythmic RGC activity and the LFPs
Rhythmic activity was recorded in the majority of RGCs in the photoreceptor degenerated retinas investigated. Our current spike sorting and cell assignment algorithm allows for the identification of ∼100 RGCs/mm2. The wt retina comprises ∼2–3000 RGCs/mm2 (Dräger and Olsen, 1981) that can be classified in ∼20 RGC subtypes (Völgyi et al., 2009). This results in ∼100–150 RGCs of each type per square millimeter. In adult rd retinas, the RGC number may remain constant (Mazzoni et al., 2008) or may decrease by ∼20% (Grafstein et al., 1972) to ∼80–100 RGCs/mm2 per subtype. It is therefore not possible to infer from our extracellular recordings whether all RGCs subtypes display rhythmic spiking behavior. However, it is unlikely that only one cell type is recorded here, as multisensor arrays detect activity from ON and OFF RGCs in mouse or rabbit retinas with equal probability (Zeck and Masland, 2007; Stasheff, 2008). Rhythmic RGC activity has been found in three morphologically identified rd1 RGCs (Margolis et al., 2008) and at much longer timescales in rod-degenerated rat RGCs (Sauvé et al., 2001; Marc et al., 2003; Sekirnjak et al., 2009; Kolomiets et al., 2010). It is therefore tempting to speculate that photoreceptor degeneration induces pathologic rhythmic activity in the human disease of retinitis pigmentosa.
Photoreceptor-degenerated retinas retain functional synaptic connections with higher visual centers in mice (Bi et al., 2006; Lagali et al., 2008; Lin et al., 2008; Thyagarajan et al., 2010) and humans (Chen et al., 2006; Zrenner et al., 2010). An early study in anesthetized rd1 mice reports rhythmic activity in neurons of the superior colliculus (Dräger and Hubel, 1978) that are postsynaptic to RGCs. Correlated RGC activity in restricted retinal areas driven by extended LFPs (Figs. 3b, 4, 5) may thus give rise to photopsias reported by RP patients. A high percentage of RP patients describe the sensation of phosphenes or some forms of flashes (Heckenlively et al., 1988; Bittner et al., 2009). We hypothesize the occurrence of phosphenes from the high percentage of synchronous RGC spikes (Fig. 2) that may lead to elevated activity in higher visual areas (Dräger and Hubel, 1978).
Maturation of neuronal circuits is characterized by spontaneous synchronized activity that propagates across extended neural circuits (Wong, 1999; Blankenship et al., 2009). A characteristic feature of developing retinas is concerted RGC activity, reflected as calcium waves or RGC population bursts (Wong, 1999; Blankenship et al., 2009). In developmental waves and in rd1 retinas (Fig. 3), consecutive population bursts are initiated in close proximity. However, in developing retinas, the interburst intervals (seconds to minutes) are much longer compared with the degenerated rd1 retinas (∼100–150 ms). We therefore compare the degenerated retinal circuit with other adult neural circuits exhibiting pathological activity. Perturbations of mature neuronal circuits (i.e., by reducing inhibition) results in aberrant oscillations, hypersynchrony, and eventually the emergence of epileptiform activity (Traub et al., 1993; Pinto et al., 2005). In the rd1 retinas, aberrant activity was detected without perturbing the circuit. However, reducing glutamatergic input abolished LFPs and rhythmic spiking (Fig. 7), suggesting that inhibitory retinal circuits may not be strong enough to compensate the excitatory driving force. Additional disinhibition of the rd1 retinas increased the synchronous discharge of RGCs (Fig. 6; supplemental Fig. 2, available at www.jneurosci.org as supplemental material). Disinhibited neural circuits in the motor cortex generate ∼10 Hz afterdischarge oscillations after each ictal spike (Castro-Alamancos et al., 2007). Afterdischarge oscillations in the neocortex (Flint and Connors, 1996) or in disinhibited hippocampal slices (Traub et al., 1993) occur at a similar frequency. Synchronous discharges in these examples rely on the activation of ionotrophic glutamate receptors, and are reflected by strong local field potentials, similar to the results obtained here (Figs. 3, 7).
In rd1 retinas, LFPs expanded and propagated across the retinal ganglion cell layer, depolarizing the RGCs in the corresponding areas. The propagation velocity (8 mm/s) is by an order of magnitude higher than the velocities measured across developmental retinal waves in different species (0.1–1.5 mm/s) (Wong, 1999). The fastest developmental waves (stage I) were abolished by gap junction blockers (Syed et al., 2004) as were the wave-like LFPs in this study. Slower waves (stage III) are suggested to rely on glutamate spillover (Blankenship et al., 2009). This mechanism appears unlikely here. The application of AMPA/kainate antagonists did not block stage III waves, indicating that extrasynaptic glutamate receptors are involved. In the rd1 retinas, LFPs are abolished by AMPA/kainate antagonists.
A rigorous comparison of rd1 LFP propagation speed with various epilepsy models cannot be performed here. However, we mention that the propagation velocity in an epileptiform model that relies on nonsynaptic coupling (Sinha and Saggau, 2001) is in the same range (10 mm/s) as the velocity measured here. This brief comparison suggests that aberrant activity in degenerated rd1 retina shares characteristic features with certain forms of epileptiform activity. However, we note differences between the two disorders. In degenerated rd1 retinas, NMDA receptors and GABAA receptors (supplemental Fig. 2, available at www.jneurosci.org as supplemental material) seem to play a minor role, in contrast to many forms of epilepsy (Flint and Connors, 1996; Castro-Alamancos et al., 2007).
Mechanisms underlying the rhythmic RGC activity and the propagating local field potentials
RGC rhythmic activity could originate within the cells themselves or could be driven by presynaptic input. A previous patch-clamp study established that three identified RGC subtypes are driven by rhythmic presynaptic input (Margolis et al., 2008). Our data are consistent with this result. Rhythmic spiking ceases after the application of iGlu receptors antagonists that suppress excitatory presynaptic input to RGCs (Fig. 7). We therefore infer that RGCs do not act as pacemakers in the rd1 retina. In the adult rd1 retina, glutamate is released by bipolar cells, as the majority of photoreceptors are lost. Recent evidence suggests that bipolar cells may generate a 10 Hz rhythm in rd1 mouse bipolar cells (Borowska et al., 2010). Indirect evidence from other species [tiger salamander (Gao et al., 2009); goldfish (Protti et al., 2000)] support this hypothesis. Other pacemaker-like interneurons have been reported in the wt retina, such as the starburst amacrine cells (Petit-Jacques et al., 2005) or dopaminergic amacrine cells (Feigenspan et al., 1998). The rhythmic activity of these cells, however, is abolished by TTX, whereas the rhythmic rd1 LFPs persist under TTX application. Our results do not exclude that inhibitory interneurons act as rhythm generators (Vaithianathan and Sagdullaev, 2010; Margolis and Detwiler, 2011). However, the pharmacological experiments (Figs. 6, 7; supplemental Fig. 2, available at www.jneurosci.org as supplemental material) demonstrate that rd1 RGCs receive an excitatory rhythmic driving force that is absent in wt retinas. This driving force is modulated but does not rely on inhibitory neurotransmitters.
How does the rhythmic activity propagate across the retina? We discussed that glutamate spillover is unlikely to mediate the activity propagation in rd1 retinas. Our results suggest that gap junctions are involved (Fig. 8). Electrical coupling is abundant in the retina (for review, see Söhl et al., 2005; Bloomfield and Völgyi, 2009) and has been reported between all major neuronal classes. Our recordings demonstrate that the coupling strength between rd1 RGCs is twice as large as that in wt retinas (Fig. 1). Strong electrical RGC coupling persists when iGlu receptors and thus LFPs are blocked (Fig. 7). However, the propagation of membrane depolarizations through strong electrical RGC synapses appears unlikely. The majority of rhythmic RGCs are out of phase (Fig. 2) and do not display electrical coupling (Fig. 1c). Photoreceptors mostly disappeared in the adult rd1 retina (Lin et al., 2009). Horizontal cells are electrically coupled in wt retinas (Hombach et al., 2004; Bloomfield and Völgyi, 2009), but their coverage diminishes during degeneration (Strettoi and Pignatelli, 2000). Among the amacrine cells, one of the most abundant cell types is the glycinergic AII cell (Wässle, 2004). Strychnine strongly modulates the LFPs in rd1 retinas (supplemental Fig. 2, available at www.jneurosci.org as supplemental material), suggesting the contribution of glycinergic cells. AII cells are electrically coupled to ON bipolar cells (Strettoi et al., 1992; Feigenspan et al., 2001) and to neighboring AIIs (Veruki and Hartveit, 2002). Signal transmission between electrically coupled neurons and specifically between AII amacrine cells exhibit low-pass filter characteristics (Veruki and Hartveit, 2002). Lower coupling strength between AII cells results in higher phase shift between paired voltage traces (Veruki et al., 2008). The almost random distribution of CC phase lags between rhythmic RGCs (Fig. 2) may be attributed to variable gap junction coupling between AII cells or between AII and bipolar cells. Indirect evidence reporting the accumulation of glycine in bipolar cells of rod-degenerated rat retinas suggests increased gap junction permeability between bipolar and AII amacrine cells (Fletcher, 2000). Furthermore, TTX inhibits voltage-gated sodium channels in AII cells and slows down the output of the AII network (Tian et al., 2010). This could explain the decreased LFP propagation velocity in TTX-treated rd1 retinas (Fig. 5, Table 1). These arguments suggest the transmission of excitation from rhythmic bipolar cells through a network of electrically coupled amacrine cells. The proposed scenario involving the AII network remains hypothetical; other types of electrically coupled and TTX-sensitive interneurons could mediate the LFP propagation.
An additional experiment is needed to elucidate the transition from normal physiological network dynamics in young, seeing rd1 retinas to the aberrant pathological activity reported here. If the rd1 pathology transfers to the human disease of retinitis pigmentosa, then therapies may aim to reduce aberrant RGC activity by interfering with the increased electrical coupling. The activation of (light-sensitive) rd1 RGCs alone is not sufficient to restore visual percept (Thyagarajan et al., 2010).
Footnotes
This work was supported by The Max Planck Society and by German Science Foundation Grant ZE 535/4-1 (G.Z.). We thank A. Lambacher for technical assistance with the spike-sorting algorithm and D. Ng for critical reading of this manuscript.
- Correspondence should be addressed to Günther Zeck at the above address. zeck{at}neuro.mpg.de or guenther.zeck{at}nmi.de