Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2009 Jun 26.
Published in final edited form as: Neuron. 2008 Jun 26;58(6):851–858. doi: 10.1016/j.neuron.2008.04.025

A precisely timed asynchronous pattern of ON and OFF retinal ganglion cell activity during propagation of retinal waves

Daniel Kerschensteiner 1, Rachel OL Wong 1
PMCID: PMC2553397  NIHMSID: NIHMS57657  PMID: 18579076

Summary

Patterns of coordinated spontaneous activity have been proposed to guide circuit refinement in many parts of the developing nervous system. It is unclear, however, how such patterns, which are thought to indiscriminately synchronize nearby cells, could provide the cues necessary to segregate functionally distinct circuits within overlapping cell populations. Here we report that glutamatergic retinal waves possess a novel substructure in the bursting of neighboring retinal ganglion cells with opposite light responses (ON or OFF). Within a wave, cells fire repetitive non-overlapping bursts in a fixed order: ON before OFF. This pattern is absent from cholinergic waves, which precede glutamate-dependent activity, providing a developmental sequence of distinct activity-encoded cues. Asynchronous bursting of ON and OFF retinal ganglion cells depends on inhibition between these parallel pathways. Similar asynchronous activity patterns could arise throughout the nervous system as inhibition matures and might help to separate connections of functionally distinct subnetworks.

Introduction

The connectivity patterns of many early neuronal networks undergo extensive refinement during development. In particular, many target neurons initially receive exuberant connections from diverse and often functionally distinct presynaptic cells and with maturation lose inputs from inappropriate synaptic partners while strengthening appropriate connections (Wong and Lichtman, 2003). Molecular cues are thought to govern initial circuit formation. In many cases, the subsequent refinement of synaptic connections appears to be guided by activity-dependent learning rules (Cline, 2003; Zhang and Poo, 2001). Since Hebb’s original conjecture (Hebb, 1949), many related plasticity rules have been proposed (Dan and Poo, 2006). While the details of these rules differ, they commonly predict the strengthening of connections between synchronously active cells and the weakening and elimination of connections between asynchronously active cells. In recent years, increasingly elaborate paired recordings have allowed for direct testing of synaptic plasticity rules in mature and developing neural circuits (Dan and Poo, 2006). These studies confirmed the importance of synchronous and asynchronous activity in modifying synaptic efficacy and revealed that the time windows during which pre- and postsynaptic activity interact can vary from millisecond to seconds depending on the dominant activity patterns: spikes and bursts of spikes, respectively (Butts et al., 2007; Kobayashi and Poo, 2004; Shah and Crair, 2008; Sjostrom et al., 2001). In addition, synaptic remodeling in some circuits was shown to be sensitive to the order of pre and postsynaptic action potentials: strengthening inputs firing before and weakening inputs firing after postsynaptic action potentials (Levy and Steward, 1983; Markram et al., 1997). In a crucial extension of these findings, in vivo studies that imposed varying activity patterns on retinal ganglion cells (RGCs) converging onto neurons in the tectum of Xenopus tadpoles demonstrated that activity-instructed segregation of convergent inputs requires repetitive, precisely-timed asynchronous firing of presynaptic cells (Zhang et al., 1998). Although many developing neural networks are known to spontaneously generate patterns of synchronized activity (Feller, 1999), apart from the partially disjoint firing of motorneurons innervating flexor and extensor muscles in embryonic chick (O'Donovan and Landmesser, 1987), no accurately timed asynchronous activity patterns have been identified.

Retinal waves are the best-studied example of spontaneously generated patterned activity (Wong, 1999). In early postnatal development, a network of cholinergic amacrine cells in the inner retina supports the slowly spreading excitation of RGCs correlating the activity of cells in a distance-dependent fashion. Cholinergic waves are required for the normal retinotopic refinement of RGC axons in both the superior colliculus (SC) and dorsolateral geniculate nucleus (dLGN) (Chandrasekaran et al., 2005; Grubb et al., 2003; McLaughlin et al., 2003), as well as for the precise mapping of geniculocortical projections (Cang et al., 2005). Prior to eye-opening, when retinal waves are driven by glutamatergic transmission, retinogeniculate projections undergo a further stage of activity-dependent refinement that separates inputs from adjacent ON and OFF RGCs (Dubin et al., 1986), which respond to light increments and decrements, respectively. At the same time, distance-dependent correlations imposed by wave propagation are thought to maintain newly established eye-specific retinotopic maps (Chapman, 2000; Demas et al., 2006). This raises the question of whether glutamatergic waves encode cues that help segregate inputs from functionally distinct neighboring cells while maintaining the overall pattern of wave propagation.

Here, using high-density multielectrode arrays, we discovered that the activity of ensembles of neighboring ON and OFF mouse RGCs during glutamatergic waves is precisely coordinated. Within each wave, cell pairs of the same sign (ON and ON, OFF and OFF) display repetitive coincident bursts of action potentials, whereas opposite-signed (ON and OFF) cell pairs fire adjacent non-overlapping bursts of action potentials in a fixed temporal order: ON before OFF. Based on pharmacological experiments we propose that this pattern, which seems to match burst-dependent plasticity rules that guide circuit refinement in the dLGN and SC (Butts et al., 2007; Shah and Crair, 2008), is generated by inhibition between these parallel pathways and a propensity of OFF RGCs to fire upon disinhibition (Margolis and Detwiler, 2007).

Results

To analyze the structure of the spontaneous activity of neighboring ON and OFF RGCs during glutamatergic waves, we recorded spikes from these cells at postnatal day 12 (P12) using planar multielectrode arrays of the design illustrated in Figure 1A. Arrays consisted of two separate recording fields 500 µm apart each containing 30 electrodes within a rectangular area 150 µm by 180 µm wide. This arrangement meant that all electrodes of one recording field were contained within an area smaller than the dendritic fields of most RGCs at this age (Diao et al., 2004), and allowed us to record the activity of ensembles of neighboring RGCs with overlapping or adjacent dendrites which were recruited nearly simultaneously into passing waves. Comparing the average activity from both recording fields revealed the propagation of activity between both electrode patches that is characteristic of retinal waves (Figure 1B).

Figure 1. Spontaneous activity of ON and OFF RGCs during glutamatergic waves.

Figure 1

Open in a new tab

(A) Illustration of the multielectrode array design. Two rectangular recording fields positioned 500 µm apart, each consisting of 30 electrodes (10 µm electrode diameter). The distance between electrodes within each recording field was 30 µm.

(B) Histograms of the average firing rate of all sorted units within a recording field. Color code indicates activity from the respective electrode patches shown in (A).

(C) Spike rasters of spontaneous activity from three ON RGCs and three OFF RGCs recorded simultaneously within one recording field. Note that some weaker waves preferentially recruit OFF RGCs.

(D) An excerpt of the same spike rasters shown in (C) presented on a finer time scale reveals unique burst structure of ON and OFF RGCs.

(E) Peristimulus rasters and histograms of spike trains from representative ON and OFF RGCs during 18 cycles of a full-field stimulus square wave modulated at 0.125 Hz. Shaded areas indicate periods of darkness (~101 Rh*/M-cone/s) and unshaded areas indicate periods of illumination (~105 Rh*/M-cone/s).

In order to identify ON and OFF RGCs, we presented a square wave modulated (0.125 Hz) full-field stimulus after recording spontaneous activity for > 1 hr in complete darkness (Figure 1E). We observed robust light responses in isolated retinas at P12, ~3 days before eye-opening in mice. We classified RGCs as ON or OFF responsive if > 80 % of their spikes occurred within the respective phase of the stimulus and as ON-OFF responsive otherwise. Thus, ~93 % of RGCs (519 of 556 RGCs recorded from 39 retinas) were either ON or OFF responsive and ~7 % were ON-OFF responsive (37 of 556 RGCs).

Most action potentials of spontaneously active cells at P12 occurred in bursts (81 ± 1 %, mean ± SEM) that lasted on average 0.61 ± 0.02 s. Similar to recordings of spontaneous glutamatergic activity from ferrets (Lee et al., 2002), we found that OFF RGCs had higher mean firing rates (0.82 ± 0.06 Hz) than ON RGCs (0.48 ± 0.04 Hz, p < 0.0001) over the course of an experiment. Interestingly, this was not due to a higher firing rate of OFF compared to ON cells within waves (OFF: 18.1 ± 0.7 Hz, ON: 20.6 ± 1.2 Hz, p > 0.4), but rather to the more frequent recruitment of OFF cells into waves of similar length (Figure 1C, p > 0.1). Consistent with this idea, the fraction of time spent participating in waves was higher for OFF than for ON cells (OFF: 8.1 ± 0.7 %, ON: 4.8 ± 0.4 %, p < 0.0001) and the interval between waves was shorter (OFF: 56 ± 6 s, ON: 127 ± 10 s, p < 0.0001).

Neighboring ON and OFF RGCs fire in precisely timed asynchronous bursts during glutamatergic waves

At P12, RGCs fired multiple bursts of action potentials during a wave. Grouping representative spike trains recorded from the same patch of electrodes according to ON or OFF responsiveness of the respective cells, revealed a striking activity pattern (Figure 1D). Bursts of cells of the same sign appeared to occur together, whereas those of opposite sign did not. In addition, the bursts of ON cells seemed to precede the bursts of OFF cells in stereotypic fashion. To quantify these observations, we computed the cross-correlations for spike trains of RGC pairs recorded from the same patch of electrodes. Figure 2A shows representative traces for all combinations of cell pairs from one experiment. The cross-correlation plots of same sign and opposite sign cell pairs confirmed that the burst times of same sign cell pairs were synchronized and those of opposite sign pairs offset such that OFF cells were most likely to fire action potentials ~1 s after the spiking of neighboring ON cells (see also Figure S1). The remarkable precision and stability of this pattern is illustrated by histograms of the peak time in the cross-correlations of all cell pairs analyzed (Figure 2B, 901 pairs, n = 22 retinas). ON and OFF cells thus display precisely coordinated firing patterns during glutamatergic retinal waves firing adjacent non-overlapping bursts of action potentials in a fixed temporal sequence: ON before OFF.

Figure 2. Precise cross-correlation structure of ON and OFF RGC spiking during glutamatergic waves.

Figure 2

Open in a new tab

(A) Cross-correlations for all combinations of spike trains of spontaneously active RGCs within one electrode field recorded in one experiment at P12. Firing pattern of ON relative to OFF RGCs (left panel), same sign RGCs (middle panel), and OFF relative to ON RGCs (right panel) are compared.

(B) Histograms representing the peak times in cross-correlations of all combinations of cell pairs (n = 22 retinas) recorded within the same electrode field. Histograms were fit with Gaussian distributions. ON to OFF RGCs (left panel, mean ± SD, −1.01 ± 0.03 s), same sign RGCs (middle panel, 0.01 ± 0.01 s) and OFF to ON RGCs (right panel, 1.03 ± 0.03 s)

Asynchronous bursting of ON and OFF RGCs is restricted in development to the period of glutamatergic waves

Glutamatergic waves in development are preceded by correlated activity that is supported by a network of cholinergic amacrine cells (Feller, 1999; Wong, 1999). These cholinergic waves have convincingly been shown to help refine retinotopic maps in subcortical and cortical visual areas while being dispensable for ON/OFF segregation (Cang et al., 2005; Chandrasekaran et al., 2005; Grubb et al., 2003; McLaughlin et al., 2003). We hypothesized therefore, that the disjoint burst pattern described above might be absent from cholinergic waves.

To analyze the development of burst patterns we compared spontaneous activity of RGCs at P10 (cholinergic waves), P12 (glutamatergic waves), and P15 (eye-opening, glutamatergic waves in decline) (Bansal et al., 2000; Demas et al., 2003). At P10, we did not observe reliable light responses and thus were unable to determine if cell pairs consisted of opposite or same sign cells. However, when all cell pairs recorded at P10 were plotted according to the peak time and amplitude of their cross-correlation function they formed a single broad cluster without clear substructure (Figure 3A). Likewise, correlation coefficients were on average positive (0.095 ± 0.006, mean ± SEM, n = 317) and their histogram for all cell pairs appeared to outline a single distribution (Figure 3B). To verify this impression, we fit both the histogram of correlation coefficients and peak times with a Gaussian mixture model with an increasing number of components and calculated Akaike and Bayesian information criteria (McLachlan and Peel, 2000). In both cases allowing multiple components failed to reduce these criteria for model selection, arguing that activity correlations of nearby RGCs during cholinergic waves were not distinguished by their response type. In contrast, at P12 spike trains from both same and opposite sign RGC pairs had high cross-correlation amplitudes (same: 0.26 ± 0.006, n = 415; opposite: 0.18 ± 0.005, n = 486) with clearly offset peak times, confirming that the activity pattern of ON and OFF RGCs during glutamatergic waves is temporally precise with ON bursts preceding OFF bursts. In addition, correlation coefficients for spike trains were negative for opposite sign pairs and positive for same sign pairs (opposite sign: −0.045 ± 0.002; same sign: 0.23 ± 0.007, p < 0.0001). Using the same statistical criteria as above to select the number of components in Gaussian mixture models verified that the histogram of peak times consisted of three populations (ON-OFF, same sign and OFF-ON pairs) and the histogram of correlation coefficients of two (opposite and same sign). By P15, activity patterns of neighboring opposite sign RGCs became poorly correlated (correlation coefficient: −0.007 ± 0.002, n = 187, p < 0.0001 for comparison to P12) and positive correlation for same sign pairs were reduced (0.097 ± 0.007, n = 211, p < 0.0001). This was matched by a decline in waves. While ~ 80 % of spikes occurred during waves both at P10 (85 ± 1%) and at P12 (80 ± 1%), only ~40 % of spikes occurred during waves at P15 (40 ± 3%). In conclusion, the precisely-timed asynchronous burst pattern of ON and OFF RGCs appears to be restricted to glutamatergic waves, and its expression declines after eye-opening as waves begin to disappear.

Figure 3. Development of the cross-correlation structure of spontaneous activity of ON and OFF RGCs.

Figure 3

Open in a new tab

(A) Each dot represents a pair of cells recorded simultaneously within one recording field at P10 (left panel, 317 cell pairs, n = 4 retinas), P12 (middle panel, 901 cell pairs, n = 22 retinas) and P15 (right panel, 398 cell pairs, n = 5 retinas). Values along the x-axis indicate peak time of cross-correlation functions. Values along the y-axis represent the value of the cross-correlation at its peak. At P10 (left panel) no reliable light responses were observed and all cell pairs are therefore depicted as grey dots. At P12 and P15, same sign cell pairs are shown as black dots and opposite sign cell pairs as purple or red dots depending on the orientation of their pairing (purple: ON – OFF, red: OFF – ON).

(B) Histograms of the correlation coefficients (i.e. value of the cross-correlation function at zero time-lag), at P10 (left panel), P12 (middle panel) and P15 (right panel). Color coding as in (A).

Inhibitory transmission desynchronizes burst times of neighboring ON and OFF RGCs during glutamatergic waves

We performed a series of pharmacological experiments to gain insight into the circuit mechanisms that underlie the distinct burst pattern of ON and OFF RGCs during glutamatergic waves. In order to facilitate the presentation of pharmacological effects on the cross-correlation structure of cell pairs, we defined an index of burst preference (BPI, see Experimental Procedures). The BPI is positive for cell pairs that fired more coincident than adjacent bursts and negative for pairs that preferentially fired adjacent bursts.

The mechanisms that shape the defining features of the RGC burst pattern during glutamatergic waves can be addressed by asking two questions: 1) What causes neighboring RGCs of the same response sign to burst synchronously, and 2) what causes neighboring ON and OFF RGCs to burst in sequence? For same sign cell pairs, blockade of electrical coupling by application of carbenoxolone or meclofenamic acid (Pan et al., 2007) reduced narrow correlations (≤ 50ms), but did not change the overall burst pattern (Figure S2). Likewise blockade of GABAergic (Figure 4B) and/or glycinergic transmission (Figure 4C and D) did not affect the burst pattern of same sign cell pairs, arguing that their coincident bursting was primarily caused by shared or synchronized excitation from presynaptic bipolar cells. This distinct influence of electrical coupling and bipolar cell input on narrow and broad correlations, respectively, is similar to what was observed previously for correlated activity in the adult retina (Brivanlou et al., 1998).

Figure 4. Synaptic inhibition underlies offset burst pattern of spontaneously active ON and OFF RGCs.

Figure 4

Open in a new tab

(A – D) Burst preference index (BPI) for same sign (black dot) and opposite sign (red dot) cell pairs recorded at P12. BPI in control conditions is shown along the x-axis. BPI in the presence of the following pharmacological agents is shown along the y-axis: (A) 50 µM L-APB (L-2-amino-4-phosphonobutyric acid), (B) 5 µM gabazine (SR95531), (C) 500 nM strychnine, and (D) 50 µM TPMPA, 5 µM gabazine and 500 nM strychnine. Based on previous studies all concentrations of blockers are assumed to be saturating. (E) Schematic of the mammalian retina. Cells of ON circuits are highlighted in white: RBC, rod bipolar cell; ON CBC, ON cone bipolar cell; ON, ON RGC. Cells of OFF circuits are highlighted in gray: OFF CBC, OFF cone bipolar cell; OFF, OFF RGCs. In addition, a glycinergic (Gly) AII amacrine cell (AII) is shown in blue. The yellow cell represents small-field diffusely stratified amacrine cells that are likely glycinergic (Gly) and mid- to wide-field diffusely or bi-stratified amacrine cells that are likely GABAergic (GABA) (MacNeil and Masland, 1998; Menger et al., 1998). Red and green arrows indicate the direction of inhibitory and excitatory transmission, respectively, in the inner plexiform layer (IPL). Below the schematic, the burst pattern within one wave in control conditions and during blockade of all inhibitory transmission is shown for representative neighboring ON and OFF RGCs.

Interestingly, application of the mGluR6 agonist L-APB (50 µM), which hyperpolarizes ON but not OFF bipolar cells (Slaughter and Miller, 1981) did not affect the relative burst patterns of spontaneously active same or opposite sign RGC pairs (Figure 4A). Moreover, L-APB increased the mean rate of spontaneous ON and OFF RGC firing (Table S1), while it blocked all light-evoked ON responses as expected (Slaughter and Miller, 1981) (Figure S3). This indicates that while L-APB-induced hyperpolarization initiated at mGluR6 receptors in ON bipolar cell dendrites is sufficient to block transmission of photoreceptor signals from the outer plexiform layer, it does not block glutamate release from bipolar cell axons during waves in the inner plexiform layer.

Given the known circuitry in the inner retina (Figure 4E) and the propensity of OFF RGCs to burst upon disinhibition (Margolis and Detwiler, 2007), the precise sequence of activity in which the bursts of ON cells invariably preceded the bursts of OFF cells suggested that circuits exciting ON RGCs concomitantly suppressed OFF RGCs directly and through inhibition of OFF bipolar cells followed by postinhibitory rebound firing of OFF RGCs. This hypothesis was tested by blocking GABAergic and glycinergic transmission (Figures 4B – D). Blockade of GABAA (by gabazine), GABAC (by TPMPA), or glycine receptors (by strychnine) increased the average firing rate of RGCs (Table S1) confirming that, as in other species (Fischer et al., 1998; Sernagor et al., 2003), these transmitters are inhibitory during the period of glutamatergic waves in mice. Neither blocking GABAA (Figure 4B) nor GABAC (data not shown) receptors alone reversed the burst preference of opposite sign cell pairs. By contrast, removal of glycinergic inhibition inverted the burst preference of ~30 % of opposite sign cell pairs toward preferentially firing coincident bursts of action potentials (Figure 4C, 34 of 107, n = 6 retinas, p < 0.0001 for reduction of correlation at ± 1 s, and increase in correlation at 0 s). Finally, blocking both GABAergic and glycinergic transmission caused reversal of the burst preference for > 90% of opposite sign cell pairs (Figure 4D and E, 53 of 58, n = 5 retinas, p < 0.0001 for reduction of correlation at ± 1 s, and increase in correlation at 0 s). Taken together, these results argue that: 1) in the absence of inhibitory transmission, the excitatory drive from ON and OFF bipolar cells to RGCs is indeed synchronized and 2) in control conditions, however, during activity of ON circuits inhibition to OFF RGC exceeds excitation. The circuits that could explain this dominant inhibition of OFF RGCs are illustrated in Figure 4E. The connectivity of glycinergic AII amacrine cells might explain the prominent effect of strychnine which is necessary and in ~30% of the cases sufficient to synchronize the activity of ON and OFF RGCs. When AII amacrine cells are activated through rod bipolar cells during a period of ON circuit activity, they are predicted to further excite ON cone bipolar cells through gap junctions while at the same time both directly and indirectly inhibiting OFF RGCs (Murphy and Rieke, 2008). In addition to this unidirectional ON → OFF inhibitory pathway, several diffusely stratifying amacrine cells both glycinergic and GABAergic have been identified morphologically (MacNeil and Masland, 1998; Menger et al., 1998) and functional GABAergic cross-over inhibition has been described in both directions albeit with some bias for the ON → OFF orientation (Chen and Linsenmeier, 1989; Roska et al., 2006; Zaghloul et al., 2003). The most parsimonious explanation of the sequential bursting of ON and OFF RGCs then is that ON → OFF inhibition silences OFF RGCs during ON bursts and causes them to burst upon disinhibition at the end of ON bursts (Margolis and Detwiler, 2007).

Discussion

Retinal waves were previously known to synchronize the activity of RGCs in a distance-dependent manner and found to assist in the eye-specific segregation and retinotopic mapping of RGC afferents and geniculocortical projections (Cang et al., 2005; Chandrasekaran et al., 2005; Chapman, 2000; Grubb et al., 2003; Huberman et al., 2006; McLaughlin et al., 2003). We discovered that glutamatergic waves around eye-opening, but not earlier cholinergic waves, comprise a novel substructure. Within each passing wave, neighboring ON and OFF RGCs fire a sequence of repetitive asynchronous bursts in a fixed temporal order: ON before OFF. We propose that this wave substructure could aid the segregation of neighboring ON and OFF RGC afferents in the dLGN and could potentially help establish early orientation selectivity as well as ON and OFF domains in the primary visual cortex (V1) (Jin et al., 2008; Miller, 1994). At the same time, distance-dependent correlations imposed by wave propagation likely serve to maintain newly-established eye-specific retinotopic maps (Chapman, 2000; Demas et al., 2006).

The activity patterns of like-signed and opposite-signed RGC neighbors we observed match remarkably well the burst-based plasticity rules recently found to guide synaptic remodeling of retinal projections in developing ferret dLGN (Butts et al., 2007) and mouse SC (Shah and Crair, 2008). Unlike most forms of spike time-dependent plasticity (Dan and Poo, 2006), burst-time-dependent plasticity (BTDP) (Butts et al., 2007) is time-symmetric. Coincident bursts of RGCs and dLGN cells strengthen connections between them, whereas adjacent non-overlapping bursts, irrespective of the order in which they occur, weaken and eliminate retinogeniculate synapses. Given some initial bias in the capacity of ON or OFF inputs to drive a particular dLGN cell, their distinct bursting pattern during glutamatergic waves is predicted to drive segregation of opposite-signed convergent inputs. Consistent with this idea, in vivo studies that used electrical stimulation to impose varying activity patterns on RGCs converging onto Xenopus tectal neurons demonstrated that segregation required repetitive precisely-timed asynchronous firing of RGC inputs (Zhang et al., 1998).

The suggestion that spontaneous activity patterns could guide ON/OFF segregation in the dLGN of mice assumes that RGC axons, which at maturity innervate separate cells (Grubb et al., 2003), initially converge. The following observations support this assumption. First, ON and OFF responsive cells, which are intermingled in dLGN of wildtype mice, form clusters in mice without cholinergic and with precocious glutamatergic waves (Grubb et al., 2003). The fact that patterns of RGC activity influence the mature response type (ON or OFF) of dLGN neurons means that activity-dependent synapse remodeling is likely to eliminate connections from RGCs of one response type (ON or OFF) to dLGN cells initially receiving input from both. Second, dLGN cells during the period of glutamatergic waves eliminate all but one to three of >20 initial RGC inputs (Hooks and Chen, 2006). Since cholinergic waves, which precede glutamate-dependent activity, provide no information on the response type of RGCs (Figure 3) the possibility that all of the > 20 inputs converging onto dLGN cells at the onset of glutamatergic waves are from RGCs of the same sign seems remote.

Since the onset of burst desynchronization between neighboring ON and OFF RGCs coincides with the change from cholinergic to glutamatergic wave propagation (Bansal et al., 2000), it appears that the patterns of cholinergic and glutamatergic waves provide two distinct sets of cues for different aspects of circuit refinement: retinotopy and ON/OFF segregation, respectively. Strikingly, in both cases these cues seem to be provided in the burst pattern of RGC firing rather than in individual spikes (Butts and Rokhsar, 2001; Torborg et al., 2005 Figure 1 and Figure 2). Retinotopic refinement and ON/OFF segregation during both periods could thus use common burst-based plasticity rules (Butts et al., 2007; Shah and Crair, 2008). Several lines of evidence support the notion that the developmental sequence, common across many species (Wong, 1999), and the precise balance of cholinergic and glutamatergic waves are crucial for reliable wiring of the visual system. First, bursts of ON and OFF cells during glutamatergic waves are only precisely offset for relatively near RGCs. This activity pattern is therefore expected to most reliably segregate converging ON and OFF afferents in the dLGN after retinotopic refinement. More importantly, as mentioned briefly above, mice lacking the β2 subunit of nicotinic acetylcholine receptors (β2−/− mice), which have no cholinergic waves and early glutamatergic waves, show reduced retinotopic refinement and excessive ON/OFF segregation of retinal afferents (Chandrasekaran et al., 2005; Chandrasekaran et al., 2007; Grubb and Thompson, 2003; McLaughlin et al., 2003). In the SC, where cells normally respond to both ON and OFF stimuli, in β2−/− mice, cells are purely ON or OFF responsive (Chandrasekaran et al., 2007), and in the dLGN, where ON or OFF responsive cells are normally intermixed, cells of a given response type form clusters (Grubb et al., 2003). An intriguing question in this context is how it is in normal development that the same sequence of retinal activity patterns can lead to the segregation of ON and OFF responses in the dLGN but not the SC. A possible answer is provided by studies on synaptic remodeling in both structures in mice. Chandrasekaran et al. (2007) found that most synaptic remodeling in SC was complete by P7, well before the onset of glutamatergic waves (Bansal et al., 2000). By contrast, Hooks and Chen (2006) observed that in dLGN most synaptic remodeling occurs between P11 and P14, the period spanned by glutamatergic waves (Bansal et al., 2000). In conclusion, cholinergic and glutamatergic waves appear to provide, in a precise developmental sequence, distinct cues on retinal position and ON or OFF responsiveness. These cues are similarly provided in the burst pattern of spontaneously active RGCs and could instruct circuit refinement by a common burst-based plasticity rule (Butts et al., 2007; Shah and Crair, 2008). The influence of this process in different subcortical target areas appears to be regulated by the timing of distinct critical periods for synaptic remodeling.

We recorded reliable light responses from mouse RGCs in retinal explants during the peak period of glutamatergic waves (P12), around three days prior to eye-opening. In ferrets, at a comparable stage of development, dLGN cells were shown to respond to naturalistic visual stimuli presented through closed eyelids (Akerman et al., 2002). This raised the possibility that early visual activity might instruct ON/OFF segregation in the dLGN. Consistent with this idea, dark rearing prior to eye-opening led to an increased convergence of ON and OFF responses in ferrets (Akerman et al., 2002). By contrast, in mice, during the period of glutamatergic waves, blockade of spontaneous activity but not visual deprivation was found to delay synaptic remodeling and pruning of RGC inputs to dLGN cells (Hooks and Chen, 2006). This suggests that in mice, spontaneous, rather than visually evoked, activity is dominant in guiding normal maturation of RGC projections. Note that both the spontaneous activity patterns we discovered here and early visual responses desynchronize the activity of neighboring ON and OFF RGCs. Thus spontaneous and visually evoked activity could well instruct ON/OFF segregation together. The apparently different relative importance of these sources of activity for ON/OFF segregation in mouse and ferret dLGN could in part be due to the different structure of RGC activity during glutamatergic waves (Figure S4). While, similar to mice, the average activity of ON and OFF RGCs in ferrets diverges during glutamatergic waves (Lee et al., 2002), the spiking of neighboring ON and OFF cells remains positively correlated, if less than for like-signed neighbors, and their burst times synchronized (Lee et al., 2002). Computational modeling showed that these activity patterns could still be used to separate opposite-signed RGC afferents by a Hebbian type learning rule, but this process was unstable (Lee et al., 2002). An interesting proposal to overcome this instability was that the spatial separation of ON and OFF recipient cells into distinct sublaminae in the dLGN of ferrets, which mice like many other mammals lack (Sanderson, 1974; Schiller and Malpeli, 1978), might bias initially coarse innervation patterns sufficiently to help stabilize activity-dependent refinement. Taken together, these differences between ferrets and mice seem to suggest a balance between the activity-independent (sublamination) and activity-dependent (spontaneous and visually evoked) cues that guide ON/OFF segregation in the dLGN. Future studies on the development of other nervous system structures that show comparable inter-species differences in their functional architecture (Ohki et al., 2005) are needed to show how general this apparent evolutionary balance between the bias of initial activity-independent mapping and the precision of activity patterns that instruct their subsequent refinement is.

In addition to shaping connectivity patterns of RGC axons in subcortical visual areas, retinal waves, which can drive bursting of cells in dLGN and V1 (Hanganu et al., 2006; Mooney et al., 1996), have been proposed to influence several aspects of geniculocortical mapping including retinotopy (Cang et al., 2005), formation of ocular dominance columns, and the size of binocular receptive fields in V1 (Huberman et al., 2006). Experimental evidence of the influence of spontaneous retinal activity on cortical wiring is thus far limited to the period of cholinergic waves. Interestingly, however, models of activity-driven development of orientation selectivity in V1 require activity patterns that desynchronize the firing of neighboring cells of opposite sign, while at larger retinotopic separations ON and OFF cells should be coactive more often than cells of the same sign (Miller, 1994). The substructure of glutamatergic retinal waves that we describe here appears to fit these requirements. We show that the activity of neighboring ON and OFF cells is anticorrelated. Furthermore, because of the precise timing with which OFF bursts follow ON bursts, it is expected that for retinal separations where this delay approximates the delay caused by the propagation of a wave, ON and OFF cells would tend to be coactive. Note that we do not wish to dispute the importance of visual experience in refining orientation selectivity in V1 (Sengpiel and Kind, 2002), and future experiments are needed to determine whether glutamatergic waves assists in the establishment of early orientation selectivity in V1. Another aspect of geniculocortical mapping that the burst pattern during glutamatergic waves if it is faithfully transmitted to dLGN cells might influence is the appearance of ON and OFF domains of dLGN afferents in V1 (Jin et al., 2008).

The precisely-timed activity of ON before OFF RGCs during glutamatergic waves and the observation that bursts of these cells were synchronized when inhibitory transmission was blocked, suggested that excitatory drive to both was synchronized but that OFF RGCs were silenced by inhibition through the circuits illustrated in Figure 4E and burst upon disinhibition at the end of ON bursts effectively time-locking this sequence. What underlies the multiple non-rhythmic ON bursts during glutamatergic waves remains to be addressed by future experiments.

An interesting parallel to the more complex activity pattern generated by maturing retinal networks during glutamatergic waves, is the partially desynchronized firing of extensor and flexor motorneurons that gradually develops in chick embryos (O'Donovan and Landmesser, 1987). Taken together, these findings raise the possibility that the activity patterns generated by many early circuits in the nervous system might undergo similar changes as inhibition matures; expressing precisely-timed asynchronous patterns that could help segregate functionally distinct subnetworks throughout the nervous system.

Experimental Procedures

All experiments were carried out according to the guidelines of the Institutional Animal Care and Use Committee at the University of Washington and in compliance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals.

Recordings

Data were obtained from C57Bl/J6 wildtype mice. Details of multi-electrode recordings and spike sorting have been described previously (Kerschensteiner et al., 2008). The design of the arrays used (HD30/10-ITO-pr, Multi Channel Systems, Reutlingen, Germany) is illustrated in Figure 1 A. A monochromatic yellow organic light emitting display (OLED, eMagin, Bellevue, WA) was used to present a full-field stimulus square wave modulated at 0.125 Hz. Intensity of the stimulus was equivalent to ~105 photoisomerizations per middle-wavelength sensitive cone per second (Rh*/M-cone/s) during ON phase of the stimulus and ~101 Rh*/M-cone/s during its OFF phase. In experiments without pharmacological interventions, spontaneous activity was first recorded in complete darkness for > 1hr followed by visual stimulation. Alternatively, spontaneous activity was recorded in complete darkness in control conditions and in the presence of the respective pharmacological agents for > 1 hr each, with full-field stimuli presented at the beginning and end of the recording. We routinely subdivided both segments of spontaneous activity and analyzed several parts separately to control the stability of the recording and to avoid mistaking continuous changes in the structure of waves during the course of an experiment for pharmacological effects. On numerous occasions washout was permitted leading in all cases to recovery of waves of similar structure to those preceding the pharmacological intervention.

Data Analysis

Data were analyzed using custom software written in Matlab (Mathworks, Natick, MA). Several patterns are evident in the firing of spontaneously active RGCs at the ages recorded (P10 – P15). Spikes of a single unit tended to occur in short bursts of high frequency discharge. The presence of simultaneous or adjacent bursts of several units in a recording field and the spread of this activity between recording fields indicated the propagation of retinal waves. In addition, single units often fired several closely spaced bursts during a wave. We used the following algorithms to automatically identify bursts and waves in our multielectrode data. First, in each spike train potential bursts were designated based on a minimum number of spikes (4) and a maximum interval between them (0.4 s). The statistical significance of each burst was then evaluated using a modified Poisson surprise method (Legendy and Salcman, 1985). This computed the probability (PC) of observing the same number of spikes (C) as observed in the current burst, in a Poisson spike train of the same mean firing rate (r) in a segment equal to the duration of the current burst (T) according to:

PC=e−rT×(rT)CC! (1)

When the probability that a Poisson spike train would generate the same number of spikes as observed was < 10−4, the respective burst was accepted. Wave identification proceeded by a similar algorithm. First, potential waves were identified in each burst train based on a maximum interval between bursts (5 s). Bursts separated by less than this interval were fused together. However, as no minimum for the number of bursts per wave was defined, also single bursts passed this step and were designated possible waves. This was necessary as the number of bursts per wave varied with age and particularly at P10 waves often consisted of single bursts. The significance of each potential wave was then assessed by counting the number of spikes (C) from other units in the same recording field during the respective time interval of length T and determining the probability (PC) that Poisson spike trains of equivalent mean firing rate (r) would generate the same number of spikes according Equation (1). As above when PC < 10−4 waves were accepted. The identification of bursts was controlled by comparing the average width of bursts defined by our algorithm with that estimated from the full-width at half-maximum of the spike train autocorrelograms (Demas et al., 2003) of a random subset of cells. Differences between these measures were statistically insignificant (p > 0.1). In addition, for each recording we made sure that our algorithm identified > 90% of the bursts and waves identified by eye on a segment of the recording.

We calculated the cross-correlation for spike trains of cell pairs in the same recording field according to:

Cxy(t)={1N−tΔt×∑i−1N−tΔt(xi−〈x〉)×(yi+tΔt−〈y〉)1N∑i−1N(xi−〈x〉)2×1N∑i−1N(yi−〈y〉)2t≥0Cyx(−t)t<0 (2)

where xi and yi represent spike counts of two cells in the i-th of N time bins, <x> and <y> signify their respective average spike counts, and t the time-lag in the cross-correlation. The width of time bins (Δt) for the results shown was 50 ms. To verify that our conclusions did not depend on the choice of bin width, we analyzed cross-correlation with a range values (10 ms – 500 ms) obtaining qualitatively similar results. Because the firing rate of spontaneously active RGCs was nonstationary at the ages recorded, meaning that it was relatively high during waves and negligible in between, we determined average spike counts using a 5 s-wide sliding window (Perkel et al., 1967). This effectively removed the positive correlation otherwise added for cells that are both silent between waves. Again, we systematically varied the width of the averaging window between 2 s and 20 s to verify that our conclusion were independent of the precise choice of this parameter. Due to the length of spike trains used in this study (~1hr recordings divided into 50ms time bins) threshold for statistical significance of correlation coefficients (Cxy(0), see Equation 2) were below |Cxy(0)| = 0.01.

To demonstrate effects of pharmacological agents on the cross-correlation structure of cells within the same recording field, we defined an index of burst preference (BPI) as:

BPI=Rxy(0s)−12(Rxy(1s)+Rxy(−1s))Rxy(0s)+12(Rxy(1s)+Rxy(−1s)) (3)

where Rxy (t) is the raw cross correlation of two cells at lag t:

Rxy(t)={1N−tΔt∑i=1N−tΔtxi×yi+tΔtt≥0Ryx(−t)t<0 (4)

Variables in Equation (4) have the same meaning as in Equation (2). The BPI uses the cross-correlation at zero time-lag to measure the likelihood of two cells (x and y) to fire coincident bursts and the average correlation at −1 s and 1 s to assess the likelihood of cell×to burst before or after cell y, respectively. The use of these time-lags was based on the position of the peak in the cross-correlation of opposite sign cell pairs (Figure 2 and Figure 3) and the average burst width of RGCs in our experiments (0.61 ± 0.02s at P12). The BPI is expected to be positive for cell pairs that fire more coincident than adjacent bursts and negative for those that preferentially fire adjacent bursts. We used the raw cross-correlation to ensure that the BPI was bound between 1 and −1. It is worth mentioning that cells with perfectly synchronized activity could still have BPI = 0 if their activity was strictly periodic such that Rxy(−1) = Rxy(0) = Rxy(1). As illustrated in Figure S1 the bursting of RGCs, however, was not periodic in the vast majority of cases (36 of 39 retinas). In the few retinas (3 of 39) that showed oscillatory burst patterns the delay between bursts of a given cell was ~2 s such that peaks in the cross-correlogram at 1 s or −1s were restricted to opposite sign cell pairs (Figure S1). We use the BPI solely as a tool to compress the correlation structure of the activity of cell pairs in order to allow for easier demonstration of pharmacological effects on activity patterns in 2D plots.

Throughout this study we used either Wilcoxon-Mann-Whitney rank sum or, in case of paired samples, Wilcoxon signed rank tests to assess statistical significance of differences between groups.

Supplementary Material

01. Supplementary Figure Legends.

Table S1. Pharmacological effects on spontaneous activity of ON and OFF RGCs at P12 Average firing rates, burst durations and interwave intervals for spontaneous activity of ON and OFF RGCs in the absence or presence of the indicated pharmacological agents are shown. Entries in all cases are mean ± SEM. Statistical significance of changes in the firing of RGCs was assessed using a non-parametric sign test. Importantly, burst duration was relatively stable to pharmacological intervention allowing us to use the same index of burst preference in all cases.

Figure S1. Repetitive burst of ON and OFF RGCs during glutamatergic waves are not, in the majority of cases, oscillatory. (A and B) Cross-correlations for all combinations of spike trains of spontaneously active RGCs within one electrode field recorded in representative experiments at P12 are shown. (A) Repetitive bursts during waves in > 90% of the retinas recorded (36 of 39) were not oscillatory as indicated by the single high peak in the cross-correlogram between −5 and 5s. An interesting feature in the cross-correlograms of same sign cell pairs are the negative going troughs on either side of the central peak. This means that same sign cells which tend to burst together are less likely than expected by chance to burst adjacently. In light of burst based plasticity rules that predict weakening of connections that fire adjacent bursts (Butts et al., 2007), silence during this period might serve to avoid weakening of connections that are meant to be strengthened. (B) In a small fraction of the retinas (3 of 39) recorded, bursting of RGCs was more rhythmic as indicated by multiple peaks in the cross correlation function. Even in these cases, however, cross-correlograms of opposite sign cell pairs were asymmetric showing a higher peak for the bursting of OFF cells after ON cells. Together, these observations argue that the activity pattern of ON and OFF RGCs during glutamatergic waves is not explained by two (ON and OFF) phase shifted oscillators, but that instead the activity of OFF RGCs is locked to that of ON RGCs in precisely-timed fashion. The interval from an OFF burst to the next ON burst and the interval between bursts of each cell were in most cases variable.

Figure S2. Block of gap junctions reduces narrow correlations of same sign cell pairs but does not change overall burst pattern (A and B) Each dot represents a pair of cells recorded simultaneously within one recording field at P12 (A: 77 cell pairs, n = 2 retinas; B: 177 cell pairs, n = 2 retinas). Opposite sign cell pairs are shown in red, same sign cell pairs in black. Values along the x-axis indicate the correlation coefficient in a 50 ms time bin around zero time-lag in control conditions. Values along the y-axis represent the same parameter calculated from segments of the recordings during which gap junctions were blocked by 50 µM carbenoxolone (A) or 100 µM meclofenamic acid (MFA) (B) (Pan et al., 2007). Either way of blocking gap junctions reduced these narrow (≤ 50ms) correlations in the spiking of same sign cell pairs, most likely by uncoupling electrical connections between neighboring RGCs and possibly between RGCs and amacrine cells (Brivanlou et al., 1998). As expected, blockade of gap junctions did not consistently change the less than random probability of opposite sign cells to fire together. (C and D) Burst preference for the same cell pairs shown in A and B (black: same sign; red: opposite sign). Burst preference in control conditions is shown along the x-axis. Burst preference in the presence of 50 µM carbenoxolone (C) or 100 µM meclofenamic acid (MFA) (D) along the y-axis. Neither way of blocking gap junctions affected the overall burst pattern of ON and OFF RGCs, arguing that these broader aspects of the cross-correlation function are independent of electrical coupling.

Figure S3. APB abolishes ON light responses at P12 (A and B) Peristimulus rasters of spike trains from simultaneously recorded representative ON (upper two panels) and OFF RGCs (lower two panels) in control conditions (A) or in the presence of 50 µM L-APB (B) during 15 cycles of a full-field stimulus square wave modulated at 0.125 Hz. Shaded areas indicate periods of darkness (~101 Rh*/M-cone/s) and unshaded areas indicate periods of illumination (~105 Rh*/M-cone/s). (C) Peak spike rate (50 ms time bin) of ON (open circles, 35 cells from 5 retinas) and OFF RGCs (filled circles, 21 cells from 5 retinas) during the respective phase of the stimulus in control conditions (x-axis) and in the presence of 50 µM L-APB (y-axis). Peak spike rate of ON RGCs is reduced to 9.7 ± 2.1% of control while OFF RGC light responses are much less affected (peak spike rate 80 ± 7.1% of control). This reduction of OFF light responses by L-APB might be explained by the presence of L-APB sensitive type III metabotropic glutamate receptors mGluR7 at the axon terminals of OFF cone bipolar cells (Brandstatter et al., 1996) and mGluR8 in both the inner and outer plexiform layer (Koulen and Brandstatter, 2002). As evident in (B) ON cells in the presence of L-APB displayed bursts of activity. These bursts resembled the pattern of spontaneous activity more than light responses in the following aspects. Unlike normal light responses, bursts were not precisely time-locked to the change in stimulus phase, yet in 4 of 5 retinas were restricted to the OFF phase of the stimulus. The timing of this activity appeared synchronized within trials between neighboring cells. (D) Plots of the latency from the offset of light to the first spike during the OFF phase of 15 stimulus trials for three pairs of ON cells in the presence of L-APB. Different cell pairs are indicated by the different symbols. This confirms that there is extensive trial-to-trial variation for onset of firing of ON cells during the OFF phase, while the firing of neighboring cells remains synchronized. OFF RGCs in addition to short latency light responses, displayed similar bursts in the presence of L-APB.

In conclusion, these observations confirm that L-APB effectively blocked all light evoked ON responses in our recordings. In addition, bursts of action potentials that resemble activity patterns observed during waves appear in both ON and OFF RGCs and are clustered in but not precisely timed to the OFF phase of the stimulus. Future experiments are needed to define the mechanisms by which light responses and spontaneous activity interact in the presence of L-APB.

Figure S4. Comparison of cross-correlation structure of ON and OFF RGC spike trains during glutamatergic waves in mice and ferrets (A and B) Traces represent the average cross-correlations for all ferret RGC pairs presented in (Lee et al., 2002) (A), and for all mouse RGC pairs from one representative experiment of the present study (B). Importantly, in ferrets the temporal structure of the cross-correlograms was very similar for same and opposite sign cell pairs, and to distinguish between them one needs to take into account the amplitude of these traces. By contrast, mouse cross-correlations of same and opposite sign cell pairs have very different temporal structures. (C and D) Dots represents pairs of neighboring RGCs (black: same sign, purple: ON – OFF, red: OFF - ON) recorded simultaneously during glutamatergic waves in ferrets (C, 27 pairs, P16 – P24) and mouse (D, 901 pairs P12) Values along the x-axis indicate peak time of cross-correlation functions. Values along the y-axis represent the value of the cross-correlation at its peak. (E and F) Histograms of the correlation coefficients of the same cell pairs as in (C) and (D). Color coding as before.

Ferret RGCs were recorded using pairs of extracellular electrodes (Lee et al., 2002). At the end of each recording, RGCs were filled with Lucifer yellow and classified as ON or OFF based on their dendritic stratification pattern; the dendritic territories of these cell pairs overlapped. Mouse RGCs were recorded using high density multielectrode arrays and cells were classified as ON or OFF based on their light responses (see Experimental Procedures and Results). Cross-correlations were restricted to cells recorded on electrodes contained within an area smaller than dendritic fields of most RGCs at the ages covered in our study (Diao et al., 2004). Since both studies recorded nearby RGCs with overlapping dendritic territories, we think that the differences in the cross-correlations of RGC pairs in ferret and mice are likely to reflect species differences (see Discussion), while we concede that it remains possible that future studies sampling a larger number of RGCs in ferrets might still uncover burst patterns similar to those described here for mouse RGCs.

In ferrets neighboring ON and OFF RGCs tend to burst together, whereas in mice they burst in sequence: ON before OFF. We briefly want to relate these differences in activity patterns to the temporal structure of the plasticity rules used in two models of dLGN circuit refinement in ferrets (Butts et al., 2007; Lee et al., 2002). The model with which Lee et al. simulated ON/OFF segregation of retinogeniculate projections used the raw cross-correlation of RGC spike trains at zero time-lag to sample activity patterns of cells. The model with which Butts et al. simulated retinotopic refinement of RGC axons in the dLGN used a BTDP rule. This rule sampled activity correlations approximately over the range of time-lags spanned by cross-correlograms shown in Figure 2. It predicted the strengthening of inputs whose bursts coincided with the bursts of the postsynaptic cell, as well as weakening of inputs that burst shortly before or after it. Importantly, this BTDP rule was experimentally supported (Butts et al., 2007). The similar temporal structure of the activity patterns of same and opposite sign RGCs in ferrets likely means that irrespective of the temporal structure of the plasticity rule used other terms of the model (e.g. competition and inhibition in Lee et al., 2002) need to be relied on heavily to successfully achieve ON/OFF segregation. By contrast, it seems likely that if a rule similar to BTDP were applied to model circuit refinement in the dLGN based on the mouse data presented here, ON/OFF segregation might more stably be instructed by the temporal structure of the RGC activity pattern. It remains possible, however, that future studies of glutamatergic waves in ferrets using MEA recordings might detect similar activity patterns as in mice.

Acknowledgements

We wish to thank Drs. Stephen J. Eglen, Peter D. Lukasiewicz and Florentina Soto for critical reading of an earlier version of the manuscript and members of the Wong lab and Christopher Lee-Messer for helpful discussions. This work was supported in part by the NIH (R.W) and a grant from the Deutsche Forschungsgemeinschaft (DFG) (D.K).

Footnotes

Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

References

  1. Akerman CJ, Smyth D, Thompson ID. Visual experience before eye-opening and the development of the retinogeniculate pathway. Neuron. 2002;36:869–879. doi: 10.1016/s0896-6273(02)01010-3. [DOI] [PubMed] [Google Scholar]
  2. Bansal A, Singer JH, Hwang BJ, Xu W, Beaudet A, Feller MB. Mice lacking specific nicotinic acetylcholine receptor subunits exhibit dramatically altered spontaneous activity patterns and reveal a limited role for retinal waves in forming ON and OFF circuits in the inner retina. J Neurosci. 2000;20:7672–7681. doi: 10.1523/JNEUROSCI.20-20-07672.2000. [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Brandstatter JH, Koulen P, Kuhn R, van der Putten H, Wassle H. Compartmental localization of a metabotropic glutamate receptor (mGluR7): two different active sites at a retinal synapse. J Neurosci. 1996;16:4749–4756. doi: 10.1523/JNEUROSCI.16-15-04749.1996. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Brivanlou IH, Warland DK, Meister M. Mechanisms of concerted firing among retinal ganglion cells. Neuron. 1998;20:527–539. doi: 10.1016/s0896-6273(00)80992-7. [DOI] [PubMed] [Google Scholar]
  5. Butts DA, Kanold PO, Shatz CJ. A burst-based "Hebbian" learning rule at retinogeniculate synapses links retinal waves to activity-dependent refinement. PLoS Biol. 2007;5:e61. doi: 10.1371/journal.pbio.0050061. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Butts DA, Rokhsar DS. The information content of spontaneous retinal waves. J Neurosci. 2001;21:961–973. doi: 10.1523/JNEUROSCI.21-03-00961.2001. [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Cang J, Renteria RC, Kaneko M, Liu X, Copenhagen DR, Stryker MP. Development of precise maps in visual cortex requires patterned spontaneous activity in the retina. Neuron. 2005;48:797–809. doi: 10.1016/j.neuron.2005.09.015. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Chandrasekaran AR, Plas DT, Gonzalez E, Crair MC. Evidence for an instructive role of retinal activity in retinotopic map refinement in the superior colliculus of the mouse. J Neurosci. 2005;25:6929–6938. doi: 10.1523/JNEUROSCI.1470-05.2005. [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Chandrasekaran AR, Shah RD, Crair MC. Developmental homeostasis of mouse retinocollicular synapses. J Neurosci. 2007;27:1746–1755. doi: 10.1523/JNEUROSCI.4383-06.2007. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Chapman B. Necessity for afferent activity to maintain eye-specific segregation in ferret lateral geniculate nucleus. Science. 2000;287:2479–2482. doi: 10.1126/science.287.5462.2479. [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Chen EP, Linsenmeier RA. Centre components of cone-driven retinal ganglion cells: differential sensitivity to 2-amino-4-phosphonobutyric acid. J Physiol. 1989;419:77–93. doi: 10.1113/jphysiol.1989.sp017862. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Cline H. Sperry and Hebb: oil and vinegar? Trends Neurosci. 2003;26:655–661. doi: 10.1016/j.tins.2003.10.005. [DOI] [PubMed] [Google Scholar]
  13. Dan Y, Poo MM. Spike timing-dependent plasticity: from synapse to perception. Physiol Rev. 2006;86:1033–1048. doi: 10.1152/physrev.00030.2005. [DOI] [PubMed] [Google Scholar]
  14. Demas J, Eglen SJ, Wong RO. Developmental loss of synchronous spontaneous activity in the mouse retina is independent of visual experience. J Neurosci. 2003;23:2851–2860. doi: 10.1523/JNEUROSCI.23-07-02851.2003. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Demas J, Sagdullaev BT, Green E, Jaubert-Miazza L, McCall MA, Gregg RG, Wong RO, Guido W. Failure to maintain eye-specific segregation in nob, a mutant with abnormally patterned retinal activity. Neuron. 2006;50:247–259. doi: 10.1016/j.neuron.2006.03.033. [DOI] [PubMed] [Google Scholar]
  16. Diao L, Sun W, Deng Q, He S. Development of the mouse retina: emerging morphological diversity of the ganglion cells. J Neurobiol. 2004;61:236–249. doi: 10.1002/neu.20041. [DOI] [PubMed] [Google Scholar]
  17. Dubin MW, Stark LA, Archer SM. A role for action-potential activity in the development of neuronal connections in the kitten retinogeniculate pathway. J Neurosci. 1986;6:1021–1036. doi: 10.1523/JNEUROSCI.06-04-01021.1986. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Feller MB. Spontaneous correlated activity in developing neural circuits. Neuron. 1999;22:653–656. doi: 10.1016/s0896-6273(00)80724-2. [DOI] [PubMed] [Google Scholar]
  19. Fischer KF, Lukasiewicz PD, Wong RO. Age-dependent and cell class-specific modulation of retinal ganglion cell bursting activity by GABA. J Neurosci. 1998;18:3767–3778. doi: 10.1523/JNEUROSCI.18-10-03767.1998. [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Grubb MS, Rossi FM, Changeux JP, Thompson ID. Abnormal functional organization in the dorsal lateral geniculate nucleus of mice lacking the beta 2 subunit of the nicotinic acetylcholine receptor. Neuron. 2003;40:1161–1172. doi: 10.1016/s0896-6273(03)00789-x. [DOI] [PubMed] [Google Scholar]
  21. Grubb MS, Thompson ID. Quantitative characterization of visual response properties in the mouse dorsal lateral geniculate nucleus. J Neurophysiol. 2003;90:3594–3607. doi: 10.1152/jn.00699.2003. [DOI] [PubMed] [Google Scholar]
  22. Hanganu IL, Ben-Ari Y, Khazipov R. Retinal waves trigger spindle bursts in the neonatal rat visual cortex. J Neurosci. 2006;26:6728–6736. doi: 10.1523/JNEUROSCI.0752-06.2006. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Hebb D. The Organization of Behavior: A Neuropsychological Theory. New York: Wiley; 1949. [Google Scholar]
  24. Hooks BM, Chen C. Distinct roles for spontaneous and visual activity in remodeling of the retinogeniculate synapse. Neuron. 2006;52:281–291. doi: 10.1016/j.neuron.2006.07.007. [DOI] [PubMed] [Google Scholar]
  25. Huberman AD, Speer CM, Chapman B. Spontaneous retinal activity mediates development of ocular dominance columns and binocular receptive fields in v1. Neuron. 2006;52:247–254. doi: 10.1016/j.neuron.2006.07.028. [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Jin JZ, Weng C, Yeh CI, Gordon JA, Ruthazer ES, Stryker MP, Swadlow HA, Alonso JM. On and off domains of geniculate afferents in cat primary visual cortex. Nat Neurosci. 2008;11:88–94. doi: 10.1038/nn2029. [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Kerschensteiner D, Liu H, Cheng CW, Demas J, Cheng SH, Hui CC, Chow RL, Wong RO. Genetic control of circuit function: Vsx1 and Irx5 transcription factors regulate contrast adaptation in the mouse retina. J Neurosci. 2008;28:2342–2352. doi: 10.1523/JNEUROSCI.4784-07.2008. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Kobayashi K, Poo MM. Spike train timing-dependent associative modification of hippocampal CA3 recurrent synapses by mossy fibers. Neuron. 2004;41:445–454. doi: 10.1016/s0896-6273(03)00873-0. [DOI] [PubMed] [Google Scholar]
  29. Koulen P, Brandstatter JH. Pre- and Postsynaptic Sites of Action of mGluR8a in the mammalian retina. Invest Ophthalmol Vis Sci. 2002;43:1933–1940. [PubMed] [Google Scholar]
  30. Lee CW, Eglen SJ, Wong RO. Segregation of ON and OFF retinogeniculate connectivity directed by patterned spontaneous activity. J Neurophysiol. 2002;88:2311–2321. doi: 10.1152/jn.00372.2002. [DOI] [PubMed] [Google Scholar]
  31. Legendy CR, Salcman M. Bursts and recurrences of bursts in the spike trains of spontaneously active striate cortex neurons. J Neurophysiol. 1985;53:926–939. doi: 10.1152/jn.1985.53.4.926. [DOI] [PubMed] [Google Scholar]
  32. Levy WB, Steward O. Temporal contiguity requirements for long-term associative potentiation/depression in the hippocampus. Neuroscience. 1983;8:791–797. doi: 10.1016/0306-4522(83)90010-6. [DOI] [PubMed] [Google Scholar]
  33. MacNeil MA, Masland RH. Extreme diversity among amacrine cells: implications for function. Neuron. 1998;20:971–982. doi: 10.1016/s0896-6273(00)80478-x. [DOI] [PubMed] [Google Scholar]
  34. Margolis DJ, Detwiler PB. Different mechanisms generate maintained activity in ON and OFF retinal ganglion cells. J Neurosci. 2007;27:5994–6005. doi: 10.1523/JNEUROSCI.0130-07.2007. [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Markram H, Lubke J, Frotscher M, Sakmann B. Regulation of synaptic efficacy by coincidence of postsynaptic APs and EPSPs. Science. 1997;275:213–215. doi: 10.1126/science.275.5297.213. [DOI] [PubMed] [Google Scholar]
  36. McLachlan G, Peel D. Finite Mixture Models. New York: JohnWiley & Sons; 2000. [Google Scholar]
  37. McLaughlin T, Torborg CL, Feller MB, O'Leary DD. Retinotopic map refinement requires spontaneous retinal waves during a brief critical period of development. Neuron. 2003;40:1147–1160. doi: 10.1016/s0896-6273(03)00790-6. [DOI] [PubMed] [Google Scholar]
  38. Menger N, Pow DV, Wassle H. Glycinergic amacrine cells of the rat retina. J Comp Neurol. 1998;401:34–46. doi: 10.1002/(sici)1096-9861(19981109)401:1<34::aid-cne3>3.0.co;2-p. [DOI] [PubMed] [Google Scholar]
  39. Miller KD. A model for the development of simple cell receptive fields and the ordered arrangement of orientation columns through activity-dependent competition between ON- and OFF-center inputs. J Neurosci. 1994;14:409–441. doi: 10.1523/JNEUROSCI.14-01-00409.1994. [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. Mooney R, Penn AA, Gallego R, Shatz CJ. Thalamic relay of spontaneous retinal activity prior to vision. Neuron. 1996;17:863–874. doi: 10.1016/s0896-6273(00)80218-4. [DOI] [PubMed] [Google Scholar]
  41. Murphy GJ, Rieke F. Signals and noise in an inhibitory interneuron diverge to control activity in nearby retinal ganglion cells. Nat Neurosci. 2008 doi: 10.1038/nn2045. [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. O'Donovan MJ, Landmesser L. The development of hindlimb motor activity studied in the isolated spinal cord of the chick embryo. J Neurosci. 1987;7:3256–3264. doi: 10.1523/JNEUROSCI.07-10-03256.1987. [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Ohki K, Chung S, Ch'ng YH, Kara P, Reid RC. Functional imaging with cellular resolution reveals precise micro-architecture in visual cortex. Nature. 2005;433:597–603. doi: 10.1038/nature03274. [DOI] [PubMed] [Google Scholar]
  44. Pan F, Mills SL, Massey SC. Screening of gap junction antagonists on dye coupling in the rabbit retina. Vis Neurosci. 2007;24:609–618. doi: 10.1017/S0952523807070472. [DOI] [PMC free article] [PubMed] [Google Scholar]
  45. Perkel DH, Gerstein GL, Moore GP. Neuronal spike trains and stochastic point processes. I. The single spike train. Biophys J. 1967;7:391–418. doi: 10.1016/S0006-3495(67)86596-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
  46. Roska B, Molnar A, Werblin FS. Parallel processing in retinal ganglion cells: how integration of space-time patterns of excitation and inhibition form the spiking output. J Neurophysiol. 2006;95:3810–3822. doi: 10.1152/jn.00113.2006. [DOI] [PubMed] [Google Scholar]
  47. Sanderson KJ. Lamination of the dorsal lateral geniculate nucleus in carnivores of the weasel (Mustelidae), raccoon (Procyonidae) and fox (Canidae) families. J Comp Neurol. 1974;153:238–266. doi: 10.1002/cne.901530303. [DOI] [PubMed] [Google Scholar]
  48. Schiller PH, Malpeli JG. Functional specificity of lateral geniculate nucleus laminae of the rhesus monkey. J Neurophysiol. 1978;41:788–797. doi: 10.1152/jn.1978.41.3.788. [DOI] [PubMed] [Google Scholar]
  49. Sengpiel F, Kind PC. The role of activity in development of the visual system. Curr Biol. 2002;12:R818–R826. doi: 10.1016/s0960-9822(02)01318-0. [DOI] [PubMed] [Google Scholar]
  50. Sernagor E, Young C, Eglen SJ. Developmental modulation of retinal wave dynamics: shedding light on the GABA saga. J Neurosci. 2003;23:7621–7629. doi: 10.1523/JNEUROSCI.23-20-07621.2003. [DOI] [PMC free article] [PubMed] [Google Scholar]
  51. Shah RD, Crair MC. Retinocollicular synapse maturation and plasticity are regulated by correlated retinal waves. J Neurosci. 2008;28:292–303. doi: 10.1523/JNEUROSCI.4276-07.2008. [DOI] [PMC free article] [PubMed] [Google Scholar]
  52. Sjostrom PJ, Turrigiano GG, Nelson SB. Rate, timing, and cooperativity jointly determine cortical synaptic plasticity. Neuron. 2001;32:1149–1164. doi: 10.1016/s0896-6273(01)00542-6. [DOI] [PubMed] [Google Scholar]
  53. Slaughter MM, Miller RF. 2-amino-4-phosphonobutyric acid: a new pharmacological tool for retina research. Science. 1981;211:182–185. doi: 10.1126/science.6255566. [DOI] [PubMed] [Google Scholar]
  54. Torborg CL, Hansen KA, Feller MB. High frequency, synchronized bursting drives eye-specific segregation of retinogeniculate projections. Nat Neurosci. 2005;8:72–78. doi: 10.1038/nn1376. [DOI] [PMC free article] [PubMed] [Google Scholar]
  55. Wong RO. Retinal waves and visual system development. Annu Rev Neurosci. 1999;22:29–47. doi: 10.1146/annurev.neuro.22.1.29. [DOI] [PubMed] [Google Scholar]
  56. Wong RO, Lichtman JW. Synapse Elimination. In: Squire LR, Bloom FE, McConnell SK, Roberts JL, Spitzer NC, Zigmond MJ, editors. Fundamental Neuroscience. San Diego: Academic Press; 2003. pp. 533–554. [Google Scholar]
  57. Zaghloul KA, Boahen K, Demb JB. Different circuits for ON and OFF retinal ganglion cells cause different contrast sensitivities. J Neurosci. 2003;23:2645–2654. doi: 10.1523/JNEUROSCI.23-07-02645.2003. [DOI] [PMC free article] [PubMed] [Google Scholar]
  58. Zhang LI, Poo MM. Electrical activity and development of neural circuits. Nat Neurosci. 2001;(4 Suppl):1207–1214. doi: 10.1038/nn753. [DOI] [PubMed] [Google Scholar]
  59. Zhang LI, Tao HW, Holt CE, Harris WA, Poo M. A critical window for cooperation and competition among developing retinotectal synapses. Nature. 1998;395:37–44. doi: 10.1038/25665. [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

01. Supplementary Figure Legends.

Table S1. Pharmacological effects on spontaneous activity of ON and OFF RGCs at P12 Average firing rates, burst durations and interwave intervals for spontaneous activity of ON and OFF RGCs in the absence or presence of the indicated pharmacological agents are shown. Entries in all cases are mean ± SEM. Statistical significance of changes in the firing of RGCs was assessed using a non-parametric sign test. Importantly, burst duration was relatively stable to pharmacological intervention allowing us to use the same index of burst preference in all cases.

Figure S1. Repetitive burst of ON and OFF RGCs during glutamatergic waves are not, in the majority of cases, oscillatory. (A and B) Cross-correlations for all combinations of spike trains of spontaneously active RGCs within one electrode field recorded in representative experiments at P12 are shown. (A) Repetitive bursts during waves in > 90% of the retinas recorded (36 of 39) were not oscillatory as indicated by the single high peak in the cross-correlogram between −5 and 5s. An interesting feature in the cross-correlograms of same sign cell pairs are the negative going troughs on either side of the central peak. This means that same sign cells which tend to burst together are less likely than expected by chance to burst adjacently. In light of burst based plasticity rules that predict weakening of connections that fire adjacent bursts (Butts et al., 2007), silence during this period might serve to avoid weakening of connections that are meant to be strengthened. (B) In a small fraction of the retinas (3 of 39) recorded, bursting of RGCs was more rhythmic as indicated by multiple peaks in the cross correlation function. Even in these cases, however, cross-correlograms of opposite sign cell pairs were asymmetric showing a higher peak for the bursting of OFF cells after ON cells. Together, these observations argue that the activity pattern of ON and OFF RGCs during glutamatergic waves is not explained by two (ON and OFF) phase shifted oscillators, but that instead the activity of OFF RGCs is locked to that of ON RGCs in precisely-timed fashion. The interval from an OFF burst to the next ON burst and the interval between bursts of each cell were in most cases variable.

Figure S2. Block of gap junctions reduces narrow correlations of same sign cell pairs but does not change overall burst pattern (A and B) Each dot represents a pair of cells recorded simultaneously within one recording field at P12 (A: 77 cell pairs, n = 2 retinas; B: 177 cell pairs, n = 2 retinas). Opposite sign cell pairs are shown in red, same sign cell pairs in black. Values along the x-axis indicate the correlation coefficient in a 50 ms time bin around zero time-lag in control conditions. Values along the y-axis represent the same parameter calculated from segments of the recordings during which gap junctions were blocked by 50 µM carbenoxolone (A) or 100 µM meclofenamic acid (MFA) (B) (Pan et al., 2007). Either way of blocking gap junctions reduced these narrow (≤ 50ms) correlations in the spiking of same sign cell pairs, most likely by uncoupling electrical connections between neighboring RGCs and possibly between RGCs and amacrine cells (Brivanlou et al., 1998). As expected, blockade of gap junctions did not consistently change the less than random probability of opposite sign cells to fire together. (C and D) Burst preference for the same cell pairs shown in A and B (black: same sign; red: opposite sign). Burst preference in control conditions is shown along the x-axis. Burst preference in the presence of 50 µM carbenoxolone (C) or 100 µM meclofenamic acid (MFA) (D) along the y-axis. Neither way of blocking gap junctions affected the overall burst pattern of ON and OFF RGCs, arguing that these broader aspects of the cross-correlation function are independent of electrical coupling.

Figure S3. APB abolishes ON light responses at P12 (A and B) Peristimulus rasters of spike trains from simultaneously recorded representative ON (upper two panels) and OFF RGCs (lower two panels) in control conditions (A) or in the presence of 50 µM L-APB (B) during 15 cycles of a full-field stimulus square wave modulated at 0.125 Hz. Shaded areas indicate periods of darkness (~101 Rh*/M-cone/s) and unshaded areas indicate periods of illumination (~105 Rh*/M-cone/s). (C) Peak spike rate (50 ms time bin) of ON (open circles, 35 cells from 5 retinas) and OFF RGCs (filled circles, 21 cells from 5 retinas) during the respective phase of the stimulus in control conditions (x-axis) and in the presence of 50 µM L-APB (y-axis). Peak spike rate of ON RGCs is reduced to 9.7 ± 2.1% of control while OFF RGC light responses are much less affected (peak spike rate 80 ± 7.1% of control). This reduction of OFF light responses by L-APB might be explained by the presence of L-APB sensitive type III metabotropic glutamate receptors mGluR7 at the axon terminals of OFF cone bipolar cells (Brandstatter et al., 1996) and mGluR8 in both the inner and outer plexiform layer (Koulen and Brandstatter, 2002). As evident in (B) ON cells in the presence of L-APB displayed bursts of activity. These bursts resembled the pattern of spontaneous activity more than light responses in the following aspects. Unlike normal light responses, bursts were not precisely time-locked to the change in stimulus phase, yet in 4 of 5 retinas were restricted to the OFF phase of the stimulus. The timing of this activity appeared synchronized within trials between neighboring cells. (D) Plots of the latency from the offset of light to the first spike during the OFF phase of 15 stimulus trials for three pairs of ON cells in the presence of L-APB. Different cell pairs are indicated by the different symbols. This confirms that there is extensive trial-to-trial variation for onset of firing of ON cells during the OFF phase, while the firing of neighboring cells remains synchronized. OFF RGCs in addition to short latency light responses, displayed similar bursts in the presence of L-APB.

In conclusion, these observations confirm that L-APB effectively blocked all light evoked ON responses in our recordings. In addition, bursts of action potentials that resemble activity patterns observed during waves appear in both ON and OFF RGCs and are clustered in but not precisely timed to the OFF phase of the stimulus. Future experiments are needed to define the mechanisms by which light responses and spontaneous activity interact in the presence of L-APB.

Figure S4. Comparison of cross-correlation structure of ON and OFF RGC spike trains during glutamatergic waves in mice and ferrets (A and B) Traces represent the average cross-correlations for all ferret RGC pairs presented in (Lee et al., 2002) (A), and for all mouse RGC pairs from one representative experiment of the present study (B). Importantly, in ferrets the temporal structure of the cross-correlograms was very similar for same and opposite sign cell pairs, and to distinguish between them one needs to take into account the amplitude of these traces. By contrast, mouse cross-correlations of same and opposite sign cell pairs have very different temporal structures. (C and D) Dots represents pairs of neighboring RGCs (black: same sign, purple: ON – OFF, red: OFF - ON) recorded simultaneously during glutamatergic waves in ferrets (C, 27 pairs, P16 – P24) and mouse (D, 901 pairs P12) Values along the x-axis indicate peak time of cross-correlation functions. Values along the y-axis represent the value of the cross-correlation at its peak. (E and F) Histograms of the correlation coefficients of the same cell pairs as in (C) and (D). Color coding as before.

Ferret RGCs were recorded using pairs of extracellular electrodes (Lee et al., 2002). At the end of each recording, RGCs were filled with Lucifer yellow and classified as ON or OFF based on their dendritic stratification pattern; the dendritic territories of these cell pairs overlapped. Mouse RGCs were recorded using high density multielectrode arrays and cells were classified as ON or OFF based on their light responses (see Experimental Procedures and Results). Cross-correlations were restricted to cells recorded on electrodes contained within an area smaller than dendritic fields of most RGCs at the ages covered in our study (Diao et al., 2004). Since both studies recorded nearby RGCs with overlapping dendritic territories, we think that the differences in the cross-correlations of RGC pairs in ferret and mice are likely to reflect species differences (see Discussion), while we concede that it remains possible that future studies sampling a larger number of RGCs in ferrets might still uncover burst patterns similar to those described here for mouse RGCs.

In ferrets neighboring ON and OFF RGCs tend to burst together, whereas in mice they burst in sequence: ON before OFF. We briefly want to relate these differences in activity patterns to the temporal structure of the plasticity rules used in two models of dLGN circuit refinement in ferrets (Butts et al., 2007; Lee et al., 2002). The model with which Lee et al. simulated ON/OFF segregation of retinogeniculate projections used the raw cross-correlation of RGC spike trains at zero time-lag to sample activity patterns of cells. The model with which Butts et al. simulated retinotopic refinement of RGC axons in the dLGN used a BTDP rule. This rule sampled activity correlations approximately over the range of time-lags spanned by cross-correlograms shown in Figure 2. It predicted the strengthening of inputs whose bursts coincided with the bursts of the postsynaptic cell, as well as weakening of inputs that burst shortly before or after it. Importantly, this BTDP rule was experimentally supported (Butts et al., 2007). The similar temporal structure of the activity patterns of same and opposite sign RGCs in ferrets likely means that irrespective of the temporal structure of the plasticity rule used other terms of the model (e.g. competition and inhibition in Lee et al., 2002) need to be relied on heavily to successfully achieve ON/OFF segregation. By contrast, it seems likely that if a rule similar to BTDP were applied to model circuit refinement in the dLGN based on the mouse data presented here, ON/OFF segregation might more stably be instructed by the temporal structure of the RGC activity pattern. It remains possible, however, that future studies of glutamatergic waves in ferrets using MEA recordings might detect similar activity patterns as in mice.

NIHMS57657-supplement-01.pdf (529KB, pdf)

RESOURCES