Oscillatory and burst discharge in the apteronotid electrosensory lateral line lobe

Oscillatory and burst discharge is recognized as a component of many sensory systems and is now believed to contribute to feature detection, frequency tuning and correlative discharge patterns across wide regions of cortex (Singer, 1993; Singer and Gray, 1995; Laurent et al., 1996; Ritz and Sejnowski, 1997). Oscillatory membrane depolarizations and burst discharge have been detected in several cell types in the electrosensory system, including electroreceptors, ganglion cells, electrosensory lateral line lobe (ELL) pyramidal and granule cells, and toral giant cells (Zakon, 1986; Turner et al., 1991b, 1995; Carr and Maler, 1986; Metzner et al., 1998; R. W. Turner and L. Maler, unpublished observations). This activity is generated in relation to modulations of a weak electric organ discharge (EOD) emitted for the purpose of electrolocation or electrocommunication. The electrosensory system is ideal for investigating the cellular basis and functional significance of oscillatory discharge, given the long history of in vivo analysis of electrosensory processing (Bullock and Heiligenberg, 1986; Heiligenberg, 1991; Kramer, 1990, 1996; Moller, 1995; Metzner and Viete 1996a,b; Bell et al., 1999; Metzner, 1999; Sugawara et al., 1999; von der Emde, 1999). In this review, we will briefly summarize the evidence for oscillatory or burst discharge in the ELL gained through in vivo recordings and focus on recent advances in our understanding of the cellular basis for this activity through in vitro analyses.

The ELL receives direct input from primary afferents conveying the activity of ampullary and tuberous receptors. A pyramidal cell body layer and granule cell body layer form prominent laminae in the ELL across its medio-lateral extent (Fig. 1) (Maler, 1979; Maler et al., 1981, 1991). The ELL is further subdivided into four topographic maps or segments of electroreceptor distribution that run in a rostro-caudal direction: the medial (MS), centromedial (CMS), centrolateral (CLS) and lateral (LS) segments (Heiligenberg and Dye, 1982; Carr and Maler, 1986). Primary afferent inputs terminate on the basilar dendrite of a basilar pyramidal cell class or on the dendrites of a granular cell class I or class II, which are interneurons with cell bodies contained within the granule cell body layer (Maler et al., 1981). Ascending processes of granule cells form excitatory contact with somatic dendrites of a non-basilar pyramidal cell class through gap junction contact (Maler 1979; Maler et al., 1981) and inhibitory contacts with both basilar and non-basilar pyramidal cells (Maler and Mugnaini, 1994; see Berman and Maler, 1999). Both basilar and non-basilar pyramidal cells project apical dendrites through an overlying tractus stratum fibrosum (tSF) and subsequently branch several times as they course through the ventral molecular layer (VML) and the dorsal molecular layer (DML) (Fig. 1). Although this summary of cell types and connectivities is very restricted, it reflects our current understanding that granule and pyramidal cells are the main cell types involved in generating oscillatory discharge in the tuberous electrosensory maps.

In vivo recordings have identified several behavioral functions that incorporate ELL pyramidal cell discharge, some of which can be localized to specific sensory maps. Cells in the MS respond to ampullary receptors activated by low frequencies of exogenous electric field modulations (Zakon, 1986; Metzner and Heiligenberg, 1991) and participate in passive electrolocation (i.e. for small invertebrate prey) and electrocommunication (Metzner and Heiligenberg, 1991; Metzner, 1999). Pyramidal cells in the CMS, CLS and LS maps respond to higher frequencies of EOD modulations conveyed by tuberous electroreceptor inputs (Bastian, 1981; Bastian and Courtright, 1991; Saunders and Bastian, 1984; Shumway, 1989; Metzner and Heiligenberg, 1991). The discharge properties of pyramidal cells in vivo indicate a role in several aspects of sensory processing, including the encoding of temporal characteristics of field modulations (Bastian, 1981; Saunders and Bastian, 1984; Shumway, 1989), acting as feature detectors (Gabbiani et al., 1996; Wessel et al., 1996; Metzner et al., 1998; Gabbiani and Metzner, 1999), setting global gain control in the ELL via the activation of descending feedback inputs (deep basilar pyramidal cells; Bastian and Courtright, 1991) and participating in the generation of negative predictive images to cancel specific sensory inputs (Bastian, 1995, 1996a, b, 1999). Shumway (1989) was the first to demonstrate that pyramidal cells discharge in relation to ‘map-specific’ behavioral functions in reporting a differential frequency tuning of amplitude modulations across the CMS, CLS and LS maps (see also Metzner et al., 1998). More recently, Metzner and Juranek (1997) carried out selective lesions of ELL maps in vivo and established that cells in the CMS are necessary and sufficient for the final generation of a jamming avoidance response, while LS cells process input related to electrocommunicatory ‘chirps’ (see also Metzner, 1999).

Although oscillatory and burst discharge has only recently been examined in ELL pyramidal cells in vivo, it is now clear that ELL pyramidal cells incorporate this activity during electrosensory processing. Some of the first evidence can be traced to Bastian (1981), who published recordings of a rhythmic discharge of pyramidal cells that outlasted a 50 ms step change in EOD amplitude. Saunders and Bastian (1984) presented some of the first intracellular recordings of spike activity per se from pyramidal cells in vivo, which have several characteristics in common with the types of oscillatory discharge recorded in vitro. Similarly, the recordings from ELL pyramidal cells of Metzner and Heiligenberg (1991) provide indications of burst discharge during the encoding of beat frequencies in EOD amplitude. Bastian and Courtright (1991) also reported a higher prevalence of burst output in superficial compared with deep basilar pyramidal cells in response to step changes in EOD amplitude. Finally, burst discharges in ELL pyramidal cells of Eigenmannia have been analyzed directly and shown to have at least a role in the process of feature detection (Gabbiani et al., 1996; Wessel et al., 1996; Metzner et al., 1998; Gabbiani and Metzner, 1999).

The ELL is ideally suited to in vitro analysis because of its highly laminar structure, clear segregation of pyramidal cell somata and dendrites and distinct primary afferent and descending feedback projections. Mathieson and Maler (1988) were the first to establish an in vitro ELL slice preparation, the procedures for which have been modified and improved upon over time (Turner et al., 1994, 1996; Berman and Maler, 1999). Briefly, this procedure consists of cutting 300 –400 μm thick slices of the ELL in the transverse or longitudinal plane and maintaining them in an in vitro slice chamber in the presence of artificial cerebrospinal fluid and 95 % O²/5 % CO². This provides far greater stability for obtaining intracellular recordings and allows for direct visual guidance in the placement of stimulating and recording electrodes. In particular, the highly laminar organization of the ELL permits recording electrodes to be placed in the pyramidal cell body layer or at varying locations along the apical dendritic axis. Another advantage of analyzing pyramidal cell activity in vitro is that spontaneous or evoked discharge reflects cellular or circuit activity within the ELL uninfluenced by ongoing sensory drive from primary afferents or feedback pathways.

Intracellular recordings in the ELL slice preparation have revealed two prominent forms of oscillatory spike discharge. The first is generated by spontaneous, slow and rhythmic membrane depolarizations of 5 –15 mV that alternately drive and inhibit spike discharge (Fig. 2A,B) (Mathieson and Maler, 1988; Turner et al., 1996). The second form of oscillatory burst is generated at a much higher frequency (30 –200 Hz; Turner et al., 1994; Lemon et al., 1998) and can be observed during the repetitive spike discharge triggered by slow membrane oscillations (Fig. 2C,D). These bursts consist of 2 –7 spikes and are terminated by a distinct burst afterhyperpolarization (bAHP; arrows in Fig. 2C,D). Spontaneous fast oscillatory bursts are very similar in form and frequency to those that can be evoked by injection of direct current (compare Fig. 2C,D and E,F).

Intracellular biocytin injections into slowly oscillating cells indicate that, in the CMS, this form of oscillatory discharge is recorded primarily, if not exclusively, in non-basilar pyramidal cells (R. W. Turner, N. J. Berman and L. Maler, unpublished observations). By comparison, basilar pyramidal cells exhibit a spontaneous discharge punctuated by large inhibitory postsynaptic potentials (IPSPs) that occur over the same range of frequencies as the slow depolarizations in non-basilar pyramidal cells (R. W. Turner and L. Maler, unpublished observations). These distinctions remain to be determined for slowly oscillating cells in the CLS and LS. Slow oscillations can be recorded immediately after slice preparation, but are often detected 5–6 h after the slices have been isolated in vitro. Once elicited, burst discharge can be very stable and recorded without significant alteration for up to 3 h after the initial impalement. We believe that the slow oscillatory discharge in pyramidal cells is the result of well-coordinated spike discharge and synaptic interactions between pyramidal and granule cells (Turner et al., 1991b, 1995). The regularity and longevity of this activity further suggest that it represents a stable and normal product of the ELL circuitry in vitro. The possibility that slow oscillatory discharge can be recorded in vivo has yet to be established, although inspection of published recordings suggests that this can be achieved (Saunders and Bastian, 1984; Gabbiani et al., 1996).

Turner et al. (1996) obtained spontaneous single-unit recordings from the pyramidal cell layer in ELL slices to define the nature of oscillatory discharge across the CMS, CLS and LS maps. Distinguishing bursting from non-bursting units in an objective manner requires a rigorous stochastic analysis. Data were first screened to exclude unit recordings containing fewer than 2000 data points or those exhibiting any non-stationarity in the original time series. Bursting and non-bursting units were then identified according to the shape of interstimulus interval (ISI) histograms, the structure of a joint interval (I) return map (Iⁿ+1 versus Iⁿ), and the ISI coefficient of variation (CV) (Fig. 3). The two principal discharge patterns recorded gave rise to either a uni-modal or a bimodal ISI histogram (Fig. 3A,B). Joint interval return maps contained a cluster of points centered on the mean ISI, with bursting units generating additional side arms of longer intervals corresponding to pauses between bursts (Fig. 3C,D). Units with a bimodal ISI histogram and joint interval return map most often exhibited a CV much greater than 1 (corresponding to a highly irregular, non-Poisson distribution). The frequency threshold separating intra- and interburst ISIs was then identified on logarithmic histograms of the instantaneous frequency and used to distinguish between bursts, pauses and background spontaneous activity (see Turner et al., 1996).

This approach identified bursting discharge in 70 % of spontaneously active units, with several properties of burst output changing across the tuberous maps. Examples of time-stamped recordings of unit activity and the defined regions of burst discharge are shown in Fig. 4. These representative cases for CMS and LS units highlight the clear difference in the form of oscillatory output recorded in these two segments. A summary of the properties of burst discharge in each of the segmental maps is plotted in Fig. 5. These plots establish that the mean period of spike bursts was longest for CMS units and decreased sharply in the CLS and LS (CMS, 2.7 s; CLS, 1.2 s; LS, 1.1 s) (Fig. 5A). Mean burst duration was also greatest for CMS units and dropped significantly in the CLS and LS (CMS, 1.0 s; CLS, 0.1 s; LS, 0.05 s) (Fig. 5B). The duration of interburst intervals (i.e. pauses between bursts) was also slightly greater for CMS units (1.7 s) than for those in the CLS (1.1 s) or LS (1.1 s), although these differences were not statistically significant. These data indicate that the observed shift in burst period between segments (Fig. 5A) can be attributed primarily to reductions in the duration of the bursts compared with the intervening pauses. The large difference in burst duration across the segments was mirrored by a similar and significant drop in the mean number of spikes per burst from the CMS to the LS (CMS, 61; CLS, 8; LS, 8) (Fig. 5C). Finally, the mean frequency of spikes within bursts increased from the CMS to the LS (CMS, 90 Hz; CLS, 130 Hz; LS, 178 Hz). Plotting the mean value of intraburst spike frequencies for all units revealed a significant overlap of spike frequencies between segments, but also a gradual shift in the maximum frequency of spike discharge attained from the CMS to the LS (Fig. 5D). The coefficient of variation (CV) of ISIs and intraburst ISIs decreased from the CMS to the LS, while the CV of burst periods increased significantly from the CMS to the LS (Turner et al., 1996). Finally, the rate of spontaneous discharge, defined as the mean spike rate during the pause between bursts, was significantly higher in the CLS (86 Hz) and LS (11.3 Hz) than in the CMS (1.4 Hz).

In summary, stochastic unit analysis and intracellular recordings demonstrate several differences in the properties of slow oscillatory discharge across ELL tuberous sensory maps. Pyramidal cells in the CMS spontaneously generate very long bursts (seconds) of low-frequency spike discharge with a long oscillatory period. In contrast, CLS and LS cells discharge much shorter spike bursts (tens or hundreds of milliseconds) with high-frequency spike discharge and shorter oscillatory periods.

Extensive anatomical evidence suggests that the underlying circuitry of the ELL is equivalent across the tuberous segmental maps (Maler, 1979; Maler et al., 1981). The observed segmental differences in oscillatory discharge may, therefore, reflect physiological factors that vary across these maps (Johnston et al., 1990; Zupanc et al., 1992; Maler and Mugnaini, 1994; Tharani et al., 1996; Bottai et al., 1997). As previously stated, we believe that slow oscillations arise from reciprocal synaptic interactions between pyramidal and granule cells. As evidence, slow oscillations in pyramidal cells can be initiated by focal ejections of the GABA^A receptor blocker bicuculline in the granule cell body layer (Turner et al., 1991b). Conversely, focal ejections of tetrodotoxin in the granule cell body layer block oscillatory discharge, while similar ejections of ω-conotoxin GVIA dramatically alter the pattern of bursting in pyramidal cells. In no case is the response of pyramidal cells to current injection affected, indicating a restriction of these drugs to the granule cell body layer (Turner and Hawkes, 1992; R. W. Turner and L. Maler, unpublished observations). Repetitive ejections of glutamate in the granule cell body layer can also recruit slow oscillatory discharge (R. W. Turner and L. Maler, unpublished observations). These results argue that slow oscillatory discharge recorded in pyramidal cells requires the rhythmic discharge, if not endogenous bursting activity, of at least one class of granule cell. The ability to induce slow oscillations in pyramidal cells by bicuculline ejections in the granule cell layer further suggests that this activity is under tonic inhibitory control. If this is true, topographic maps in the ELL may function as conditional bursting circuits which can be controlled via inhibitory mechanisms acting at least at the level of the granule cell layer.

Our current understanding of fish behavior suggests at least two possible functional roles for slow oscillatory discharge in the ELL. First, Rose and Canfield (1993) demonstrated that Sternopygus macrurus can adjust its body position with respect to slow oscillations of a protective cover with a maximal gain of 0.1 –0.2 Hz. Since electric fish conceal themselves amongst vegetation or other objects under the influence of river currents during the day, it is possible that slow oscillatory discharge could act as a frequency-tuning mechanism to help position the body of the animal with respect to secure resting points. Second, slow oscillations may serve to tune the ELL circuitry to low amplitude modulation ‘beat’ frequencies arising from the superimposition of the EODs of nearby fish. The highly constant frequency of this input would lend itself well to being encoded by oscillatory frequencies intrinsic to each of the maps. In fact, Shumway (1989) reported that in the CMS, the I cell (non-basilar pyramidal cell) functions as a low-pass filter of amplitude modulations in field strength. This also fits with our finding that the lowest frequencies of oscillatory discharge are most clearly indicated in vitro for non-basilar pyramidal cells in the CMS. Slow oscillatory discharge in the CMS may then account for why this segment is preferentially involved in responding to the 1 –6 Hz amplitude modulation frequencies most suited to eliciting a jamming avoidance response (Heiligenberg, 1991; Metzner and Juranek, 1997; Metzner, 1999).

The majority of intracellular work on burst discharge in vitro has focused on the mechanisms underlying fast oscillatory spike bursts. This relates to the fact that these bursts are generated by way of a novel mechanism involving an active ‘backpropagation’ of Na+ spikes from the soma into apical dendrites (Turner et al., 1994). Spike backpropagation is now recognized in many different cell types (Turner et al., 1991a; Stuart and Sakmann, 1994; Johnston et al., 1996), but a role for backpropagating spikes in generating oscillatory discharge was first demonstrated in ELL pyramidal cells. Unlike the slow oscillatory discharge, fast oscillatory bursts can be evoked by injection of direct current and are recorded in both basilar and non-basilar pyramidal cells.

Turner et al. (1994) carried out a combined immunocytochemical and electrophysiological study to examine the distribution and function of Na+ channels over the pyramidal cell axis. Na+ channels were localized at the light and electron microscope levels using a polyclonal antibody directed against the Na+ channel from the electric organ of Electrophorus electricus (Ellisman and Levinson, 1982). Immunocytochemistry revealed a dense but non-uniform distribution of Na+ channel immunolabel over the external membrane surface of both basilar and non-basilar pyramidal cell bodies. In contrast, immunolabel was distributed in a punctate fashion over the initial 200 μm of pyramidal cell apical dendrites. Electron microscopy indicated that this label was associated with the external membrane of apical dendritic shafts, but was not distributed in any clear manner with respect to glial processes, pre-or postsynaptic membrane specializations or dendritic spines.

Intracellular recordings in the soma or proximal apical dendrites indicate that spike discharge can be evoked by depolarizing current injection at either location (Fig. 6). In both the soma and apical dendrites, spike discharge is typically tonic near threshold, with an initial lag and subsequent increase in frequency through the spike train (Mathieson and Maler, 1988; Turner et al., 1994). Somatic spikes are of large amplitude and short duration (81 ±1.9 mV; 0.35 ±0.03 ms half-width; N=20) compared with the lower-amplitude and broader-duration dendritic spikes (67.5 ±1.9 mV; 0.84 ±0.06 ms half-width; N=20; means ± S.E.M.). The duration of dendritic spikes also increases with distance from the soma, with the overall duration of dendritic spikes increasing from 2 –3 ms in the more proximal dendrites (50 μm) to over 20 ms in more distal locations (>500 μm). Somatic spikes are also followed by both a fast afterhyperpolarization (AHP) and a slow AHP, while only a slow AHP is present in the dendritic membrane (Fig. 6A,B). Antidromic spike discharge can be evoked in both the soma and apical dendrites by stimulating pyramidal cell axons in the plexiform layer. Antidromic spikes recorded from separate somatic and apical dendritic impalements are superimposed in Fig. 6C to compare the spike characteristics directly. This illustrates the clear difference in spike amplitudes and durations, and identifies an additional depolarizing afterpotential (DAP) of 2 –4 ms that follows the somatic spike. During repetitive spike discharge, this initially small DAP exerts an unusually pronounced influence on pyramidal cell output.

Current-evoked oscillatory spike discharge is directly related to both the amplitude and duration of membrane depolarization, with a distinct threshold at which pyramidal cells convert from a tonic to a burst output (Lemon et al., 1998). Fig. 7A illustrates a pyramidal cell exhibiting several periods of fast oscillatory bursts of 3 –4 spikes, with each burst terminated in a burst AHP (arrows). Close examination of this process indicates that each burst is driven by a gradual change in spike afterpotential (Fig. 7B). Specifically, the amplitudes of both the fast and slow AHPs decrease, while the DAP increases in amplitude with each spike in the burst. The net depolarization provided by the DAP eventually triggers a high-frequency spike doublet at the soma, followed by the burst AHP that terminates the burst.

Interestingly, the burst AHP also resets this process, so that spike afterpotentials recover to near their original amplitudes by the time of the next spike discharge (Fig. 7A). In this manner, spike discharge and afterpotentials generate an entirely intrinsic mechanism that drives the burst, triggers the spike doublet and burst AHP, and allows for the recovery of afterpotentials before the next burst begins.

This sequence of events emphasizes the importance of the DAP to burst generation in pyramidal cells. Given the similarity in the duration of the DAP and dendritic spike (Fig. 6C), we determined the process for spike initiation and conduction in pyramidal cells to identify the potential source of current generating the DAP (Turner et al., 1994). By using focal ejections of tetrodotoxin (TTX) in somatic and dendritic regions, we were able to determine that spike discharge in both the soma and apical dendrites is Na+-dependent. Moreover, measurements of antidromic spike conduction over the cell axis, spike latency considerations and the pattern of spike blockade by TTX indicated that the area with the lowest threshold for spike discharge was located in the cell body region (soma or axon hillock).

These results suggest a model for spike conduction over the cell axis and for DAP generation in pyramidal cell somata (Fig. 8). According to this model, membrane depolarizations (somatic or dendritic) initiate spike discharge near the soma. This is followed by an active spike backpropagation over at least the initial 200 μm of apical dendrites, characterized by a rapid increase in dendritic spike duration as it conducts away from the soma. The inward current associated with dendritic spike discharge results in a simultaneous sourcing of current back to the soma. The long time course of this current flow with respect to the short-duration somatic spike results in a DAP at the soma. Therefore, a key element to this model is the relative difference in the duration of somatic and dendritic spikes.

Fig. 9 illustrates the first test of this model. This experiment was designed to test the hypothesis that the dendritic spike is directly related to DAP generation. Plexiform layer stimulation was used to evoke an antidromic somatic spike, and the dendritic population spike was monitored via an extracellular recording electrode positioned in the VML. The extracellular recording electrode also contained 16 μmol l −1 TTX for focal pressure ejections into the dendritic region (Fig. 9A). Ejections of TTX (100 ms; 10 p.s.i.; 3 –6 pulses) could reduce the amplitude of the antidromic dendritic population spike by up to 90 % prior to any effect on somatic spike amplitude. However, there was an immediate and selective reduction in the amplitude of the DAP at the soma coincident with a reduction in the amplitude of the dendritic population spike (Fig. 9C). The difference between the control and test antidromic somatic spikes revealed a depolarizing potential of 7.3+0.66 mV amplitude and 3.0+0.32 ms duration (mean ± S.E.M., N=5), a duration similar to that of dendritic spikes.

These experiments were extended to current-evoked repetitive spike discharge where prevailing conditions allow the DAP to grow in magnitude and contribute to oscillatory bursts. A similar experimental arrangement is illustrated in Fig. 10A, with recording of current-evoked burst discharge at the soma and a TTX-containing pressure electrode positioned in the VML. Focal ejections of TTX into the proximal dendritic region again selectively blocked the DAP (Fig. 10B,C), but also converted cells exhibiting oscillatory output to tonic spike discharge (Fig. 10D,E). Since these effects were fully reversible, it was possible to repeatedly to switch a pyramidal cell between oscillatory and tonic output with successive TTX ejections. However, during the time when the DAP was blocked, oscillatory discharge could not be re-evoked even with 1.5 nA current pulses.

These studies support the proposed model in demonstrating that dendritic spike discharge is the source of current generating the somatic DAP and that the DAP directly influences the ability of pyramidal cells to generate oscillatory spike bursts. Given the importance of the relative duration of somatic and dendritic spikes to this process (a very narrow somatic spike and a broad dendritic spike), future experiments will need to determine the factors that control spike duration. It will also be important to determine the factors that change during repetitive spike discharge and allow the DAP to increase in amplitude and drive the oscillatory burst. Indeed, recent work indicates that repolarization of somatic and dendritic spikes is differentially controlled by K+ channel activity and that the rate of spike repolarization modulates DAP amplitude and oscillatory burst threshold (Turner et al., 1997; Lemon et al., 1998). A primary candidate for this is the Kv3A class of K+ channel recently shown by in situ hybridization and patch-clamp recordings to be localized to pyramidal cell dendrites and cell bodies (Morales et al., 1998; Rashid and Dunn, 1998). The relevance of this model to other central nervous system cell types was recently supported in a modeling study that analyzed the effects of backpropagating dendritic spikes upon the discharge properties of cortical pyramidal cells (Mainen and Sejnowski, 1996).

As mentioned above, spike doublets and bursts can be discerned in pyramidal cell recordings in vivo from both Eigenmannia and Apteronotus sp. (Saunders and Bastian, 1984; Gabbiani et al., 1996). One functional role for burst discharge has already been established. Since they have been shown to occur with highest probability in relation to appropriate positive or negative deflections in an electric field, Gabbiani and colleagues have proposed that bursts play a key role in feature extraction (Gabbiani et al., 1996; Metzner et al., 1998; Gabbiani and Metzner, 1999).

A second possible role for fast oscillatory bursts would be to establish an intrinsic cellular mechanism that contributes to the frequency-selectivity of pyramidal cells. Shumway (1989) described a differential sensitivity of pyramidal cells to AM modulations over a range of 1 –120 Hz. Metzner and Heiligenberg (1991) demonstrated an even greater frequency-responsiveness and differential sensitivity of pyramidal cells to single-cycle electric organ discharge (EOD) interruptions (‘chirps’) of only 2–5 ms (corresponding to an instantaneous frequency of 200–500 Hz). In fact, Metzner and Juranek (1997) have now shown, using selective lesions in vivo, that responses to chirps are lost after LS lesions (see also Metzner, 1999). Although evidence is mounting that the frequency of fast oscillatory bursts also changes across the ELL maps in vitro (Turner et al., 1996; R. W. Turner, unpublished observations), further work will be required to determine the degree of correspondence between oscillatory frequency and temporal encoding of high-frequency field modulations (Metzner et al., 1998).

The repetitive nature and high frequency of fast oscillatory bursts are most easily related to inputs encoding regular modulations in the EOD. Although speculative at this time, the most consistent of these would be the beat frequencies of EOD modulations generated when EOD waveforms of two neighboring fish superimpose. The work of Heiligenberg (1991) demonstrated conclusively that electric fish can use beat frequencies to determine the difference between the frequency of their own EOD and that of a conspecific. Bastian (1981) determined that basilar pyramidal cells responded maximally to amplitude modulations of approximately 64 Hz, although specific attention to the recording position across ELL maps was not routinely made. As previously stated, Shumway (1989) tested the sensitivity of pyramidal cells at higher frequencies and found that LS cells in Eigenmannia responded well to AM modulation frequencies of up to at least 120 Hz. Since Apteronotus leptorhynchus generate higher baseline EOD frequencies than Eigenmannia (range 700–1200 Hz; Hopkins, 1976), one can expect these animals to encounter higher beat frequencies during interactions with conspecifics. It will thus be important for future in vivo studies to examine the responsiveness of A. leptorhynchus pyramidal cells to beat frequencies above 120 Hz. It is interesting to note that fast oscillatory spike bursts in this species have been recorded in vitro at rates of up to 200 Hz (Lemon et al., 1998). This is an extremely high frequency for oscillatory discharge, but corresponds to that necessary to tune pyramidal cells to beat frequencies the animal should encounter in vivo.

In recent years, both in vivo and in vitro studies on electrosensory processing in gymnotids have revealed the occurrence of burst discharges in ELL pyramidal cells. A slow oscillatory discharge attributable to a synaptic mechanism and a fast oscillatory discharge attributable to an intrinsic mechanism have both been identified in pyramidal cells in vitro. Fast spike bursts are generated by way of a newly recognized interaction between somatic and dendritic membranes that may be effective in many other cell types. An important role for fast bursts in feature extraction of electrosensory input has already been identified. The number of potential roles for oscillatory output in the ELL will probably increase as whole-animal and slice experiments continue to uncover map-specific characteristics of sensory processing. The ability to apply such a range of recording techniques to one problem is a unique aspect of the electrosensory system, affording the realistic goal of establishing the cellular mechanism and role for burst discharge in sensory processing. This work was supported by Canada MRC grants to R. W. T. and L. M. R. W. T. is a AHFMR Senior Scholar.