Distinct neuron phenotypes may serve object feature sensing in the electrosensory lobe of Gymnotus omarorum

Mormyriformes in Africa and Gymnotiformes in America have convergently evolved an active electric sense, which appears to be their most important sensory modality (Lissmann, 1958; Fessard, 1974; Bullock and Heiligenberg, 1986; Moller, 1995; Bullock et al., 2005). These fishes exhibit neurally commanded electrogenic organs that generate electric organ discharges (EODs). In ‘pulse species’ the EODs are brief and well separated by silent intervals. In ‘wave species’ the EOD duration is similar to its repetition interval and consequently successive EODs are fused into a continuous wave (Fessard, 1974; Bullock and Heiligenberg, 1986; Moller, 1995; Bullock et al., 2005).

The role of the EOD in this active electrosensory system is to polarize the fish's surroundings. Within the polarized field, nearby objects behave as virtual electric sources according to their intensive properties, for instance their impedance relative to the water, their extensive properties, for instance their size and shape, and their positional attributes, i.e. their location and orientation relative to the fish’s coordinates (Lissmann and Machin, 1958; Heiligenberg, 1975; Rasnow, 1994; Caputi and Budelli, 2006; Nelson and MacIver, 2006; Caputi, 2017, 2020). The virtual electric field generated by a polarized object sums with the self-generated field and with the virtual fields of any other nearby objects (Migliaro et al., 2005; Pereira and Caputi, 2010).

The sempiternal repetition of the EOD produces a flow of electrosensory images on the fish’s skin, which are transduced and encoded by a mosaic of cutaneous electroreceptors tuned to the time course of the self-generated, species-specific EOD (Wright, 1958; Fessard and Szabo, 1961; Bullock et al., 1961; Hopkins, 1976; Watson and Bastian, 1979; von der Emde and Bleckmann, 1994; Caputi and Aguilera, 2019). Cutaneous electroreceptors are complex organs innervated by either one (Gymnotiformes) or a few (Mormyroidea) afferent fibers that project to the electrosensory lateral line lobe (ELL), forming a kind of ‘electrosensory retina’ (Lissmann, 1958; Bell and Maler, 2005).

In Gymnotus omarorum, two types of afferent fibers give rise to two parallel electrosensory pathways, fast and slow, that project via the ELL to the mesencephalon (Réthelyi and Szabo, 1973). The fast electrosensory pathway is represented in the ELL by a single cell type, which in turn projects to a mesencephalic Jeffress-type latency comparator circuit (Castelló et al., 1998; Nogueira et al., 2006; Nogueira and Caputi, 2011, 2013, 2014; Caputi and Nogueira, 2012; Carr and Maler, 1986; Carr et al., 1986a,b; Sotelo et al., 1975). The slow electrosensory pathway forms a cerebellum-like circuit in the ELL that compares ongoing electrosensory afferences with contextual expectations, integrating sensory and motor descending inputs (Réthelyi and Szabo, 1973).

As in other early sensory relays, the ELL of all electric fish has a dual-output system (Mormyroidea: Bell and Szabo, 1986; Bell, 1986; Bell et al., 1997; Meek et al., 1999; Mohr et al., 2003; Gymnotiformes: Réthelyi and Szabo, 1973; Maler, 1979, 2009a,b; Maler et al., 1982; Shumway, 1989; Saunders and Bastian, 1984; Krahe and Maler, 2014; Clarke et al., 2015; Rodríguez-Cattáneo et al., 2020). In general, ON neurons have thick basilar dendritic trees directly stimulated by primary afferents (Meek et al., 1999; Réthelyi and Szabo, 1973; Maler, 1979). Consequently, they increase their firing probability when a local stimulus increases at the center of their receptive field (Bell et al., 1997; Saunders and Bastian, 1984; Clarke et al., 2014). Conversely, OFF neurons receive predominantly inhibitory inputs following the EOD, driven indirectly via interneurons contacted by primary afferents from the center of their receptive field (Bell et al., 1997; Saunders and Bastian, 1984; Berman and Maler, 1999).

Few articles address the functional analysis of ELL neurons in pulse Gymnotiformes, most of them dealing with their in vivo activity (Schlegel, 1973; Stoddard, 1997; Castelló et al., 1998; Pereira et al., 2005, 2014; Rodríguez-Cattáneo et al., 2020). Only the intrinsic properties of spherical cells in the fast electrosensory pathway have been studied in vitro (Nogueira et al., 2006; Nogueira and Caputi, 2011, 2013, 2014; Caputi and Nogueira, 2012; Matsushita et al., 2012).

This article reports that basilar and non-basilar pyramidal neurons not only differ in their anatomy, but also in their intrinsic electrophysiological properties and thus are clear-cut anatomo-functional neuronal phenotypes subserving the dual processing of electrosensory signals in a complementary manner. Our data also shed light on two still-unanswered crucial questions.

The first question is suggested by in vivo unitary recordings in the ELL of G. omarorum (Pereira et al., 2014; Pereira, 2016: Rodríguez-Cattáneo, 2017; Rodríguez-Cattáneo et al., 2020). In the presence of stationary metal objects, ON units increase and OFF units decrease their firing rate in the expected characteristic way (Pereira et al., 2014). However, when a moving metal object crosses through their receptive field, OFF units show paradoxical reductions in latency and increases in firing probability (Rodríguez-Cattáneo et al., 2020). We have explored how the intrinsic properties of the non-basilar neurons contribute to this paradoxical behavior.

Another important question is how the basal activity of OFF neurons is maintained in pulse Gymnotiformes. Signals transmitted by OFF neurons to higher centers are encoded in a down modulation of their repetitive basal activity, modulated by inhibitory interneurons carrying sensory afferent information. In pulse Mormyroidea, this basal activity pattern is driven by an EOD command corollary discharge (Bell et al., 1997; Mohr et al., 2003). The mechanisms in Gymnotiformes remain unknown.

We show here that non-basilar neurons show spontaneous pacemaker activity (which can be modulated both up and down) and spike-firing adaptation, enabling them to respond better to rapid sensory signal variations. In turn, basilar neurons respond better to sustained inputs as expected for classical ON output cells.

Experiments were performed on 30 fish of the species G. omarorum (Richer-de-Forges et al., 2009; total fish length: 12–18 cm; unknown sex). Fish were gathered in lakes and creeks close to Montevideo (latitude 35.5 deg, longitude 55 deg) under the Authorization of the Ministry of Ganadería Agricultura y Pesca, Uruguay.

Experiments were approved by the Animal Ethics Committee of the Instituto de Investigaciones Biológicas Clemente Estable (protocol number 001/03/2011). Fish care and experiments were performed under the regulations of the Comisión Honoraria de Experimentación Animal of the Universidad de la República (ordinance 4332-99) and the International Guiding Principles for Biomedical Research Involving Animals (https://olaw.nih.gov/resources/tutorial/iacuc.htm). All surgical procedures were performed under deep anesthesia. For anatomical experiments initial immersion in MS-222 (0.5 g l⁻¹; Merck, E10521-50G) was followed by perfusion of the same solution via the gills. For in vitro slice preparation, fish were anesthetized rapidly by immersion in a cold solution of carbogen in water (95% O₂/5% CO₂ at 5°C) (Kugino et al., 2016) before opening the skull to remove the brain.

Golgi stains and intracellular labeling with biotinylated compounds were used to characterize the morphology of basilar and non-basilar superficial pyramidal neurons.

Golgi stains of the ELL were performed according to the double argentic impregnation of Ramón y Cajal and De Castro (1933). Ten fish were fixed with a dialdehyde mixture, containing glutaraldehyde 1% and paraformaldehyde 1%, in phosphate buffer (PB) (0.1 mol l⁻¹, pH 7.4). The fixative was perfused via the aorta, following saline perfusion of the vasculature. After fixation the brains were dissected out.

As mentioned before, the main axis of the ELL of pulse Gymnotiformes runs obliquely at 45 deg with respect to the brain midline. Parallel to this axis there are four somatotopic maps receiving the primary electrosensory afferent's projection. In these maps, the head is represented in the medial–rostral region and the tail is represented at the ELL pole located caudo-laterally (Castelló et al., 1998). These maps also receive descending inputs via parallel fibers running transversally. It has been reported in wave Gymnotiformes that pyramidal ELL neurons have apical dendritic trees flattened along a plane orthogonal to the parallel fibers, similar to the Purkinje neurons in the cerebellum (Saunders and Bastian, 1984). To better explore the spatial orientation of the dendritic trees of basilar and non-basilar pyramidal cells regarding these inputs, the brain was prepared so that the two ELLs were cut in orthogonal planes. A base for sectioning was cut at 45 deg (blue dotted line in Fig. 1A) with respect to the frontal plane (FP in Fig. 1A) and longitudinal axis (SP in Fig. 1A) of the brain. This procedure allowed us to cut the ELL on one side transversely (Fig. 1A, first to third light blue dotted lines) and on the other side parallel (Fig. 1A, fourth and fifth light blue dotted lines) to the ELL main axis, corresponding to the orientation of the electrosensory somatotopic multiple maps (the four pisciculi; Castelló et al., 1998; Heiligenberg and Dye, 1982).

Two brains were cut in coronal sections parallel to the frontal plane (FP in Fig. 1A). This made it possible to compare the neurons studied in vitro with earlier anatomical descriptions (Réthelyi and Szabo, 1973; Maler, 1979).

The prepared brainstems were then immersed in an aqueous solution of K₂Cr₂O₇ (5% in distilled water) for 1–2 days, followed by an aqueous solution of AgNO₃ (0.75%) for 1–2 days. This procedure was repeated a second time. Each whole piece was then dehydrated in increasing concentrations of ethanol, followed by acetone and embedded in Durcupan^® using a series of increasing concentrations of a Durcupan^® mixture diluted in acetone. After overnight in pure Durcupan^® mixture (no acetone), the samples were transferred to fresh pure Durcupan^® mixture and polymerized at 60°C for 2 days. The brainstems were sectioned at 100–150 µm using a sliding microtome. Preparations were observed with either a Zeiss Axioscope 50 or an Olympus IX-81 upright, bright-field microscope, and photographed at high resolution using PixeLink or Olympus microscope cameras. Selected Golgi-stained neurons were drawn by hand using a bright-field microscope equipped with a camera lucida device.

We also explored the morphological characteristics of electrophysiologically recorded individual neurons labeled by intracellular iontophoresis of biocytin (MERK B4261-50MG). After electrophysiological recording, the slices containing labeled neurons were fixed overnight in paraformaldehyde (4% in PB), rinsed in PB and incubated in an HRP–Avidin (Vector^®) conjugated solution in the presence of 0.3% Triton X-100 for 1 h. After thorough rinsing in PB, the presence of HRP was demonstrated by the local oxidation using a Vector^® SG Substrate kit (SK-4700). Slices were then counterstained with a solution of Methylene Blue (1% in distilled water), dehydrated, mounted with Canada balsam and observed as described above.

Whole-cell patch recordings were performed in 200 µm thick brainstem slices cut in the coronal plane (i.e. parallel to the frontal plane marked as FP in Fig. 1A) with a Leica (1000S) vibratome. Recordings were obtained from the slices corresponding to the anterior regions of the ELL that only included the tuberous maps. After sectioning, slices were incubated for 30–60 min in a low-sodium solution containing (in mmol l⁻¹): 2 KCl, 2.6 CaCl₂, 1.25 KHPO₄, 24 NaHCO₃, 1.6 MgSO₄, 20 glucose and 201 sucrose, bubbled with carbogen (95% O₂/5% CO₂, pH 7.4). Slices were then transferred to the standard recording solution of similar composition but in which sucrose was substituted by a physiological concentration of NaCl (120 mmol l⁻¹) where they were incubated for between 2 and 6 h until they were transferred to the recording chamber placed in a fixed-stage, transmitted light microscope (Zeiss Axioskop 2 FS).

Healthy neurons were identified in the polymorphic cell layer of the ELL (Maler, 1979) using Nomarski optics under infrared illumination. Whole-cell patch recordings were obtained using 6–12 MΩ tip-polished borosilicate micropipettes filled with an intracellular solution containing (in mmol l⁻¹) 122 potassium gluconate, 2.5 MgCl₂, 5.6 magnesium gluconate, 0.3 CaCl₂, 5 Na₂ATP, 5 K-Hepes, 5 H-Hepes and 1 EGTA (pH 7.4) containing biocytin (1%) making it possible to later morphologically identify the recorded neurons.

Intracellular voltage and stimulation currents were recorded using an Axoclamp 2B amplifier (Molecular Devices, San Jose, CA, USA). After stabilizing the membrane potential, we recorded the resting activity and responses to rectangular current pulses or to extracellular electrical stimuli in the dorsal molecular layer. In 16 neurons, we explored the responses to trains of rectangular pulses lasting 500 ms, separated by 1 or 3 s, depending on the experiment. The pulse amplitude was incremented in equal magnitude steps of 30 to 80 pA, depending on the cell. In our cells, we explored the responses to hyperpolarizing and depolarizing current ramps ranging from −400 to 1000 pA s⁻¹. All experiments were performed at room temperature (19–24°C).

Systematic analysis of time course of action potentials was performed in 19 healthy recorded neurons using Octave^®. We estimated the first-time derivative as the quotient of the increment of transmembrane voltage and the sampling time interval (ΔV/Δt).

Membrane voltage and its first derivative were plotted versus time and phase portraits were constructed by plotting the derivative versus the voltage.

The following parameters were measured: (1) the presence of spontaneous pacemaker-like pre-potentials that drive regular activity at rest; (2) the maximum rising slope; (3) the peak value; (4) the maximum falling slope; (5) the interval between the peaks of the first derivative; (6) the ratio between the peaks of the first derivative; and (7) the interval between the positive peak of the action potential and the negative peak of the hyperpolarizing after-potential. We used non-parametric tests to assess whether the neuron morphological type is associated with significant differences in the median values of each of these parameters.

Neuron responsiveness to current injection in current clamp mode was assessed using 500 ms duration steps of different amplitudes in 16 neurons. During each stimulation step, two parameters were evaluated: (1) the total number of spikes over the whole duration of the stimulus; and (2) the sequence of inter-spike intervals and the spiking instantaneous frequency (IF, defined as the inverse of the inter-spike interval). Polynomial regression was employed to evaluate spike rate (number of spikes in 500 ms) as a function of current step amplitude, and IF as a function of time and current step amplitude.

Data files and Octave^® codes are available from https://drive.google.com/file/d/1lRG7Ogm5mW6uqJVI9RN1SwpQmNgSO43T/view?usp=sharing

The ELL of gymnotiform weakly electric fish is a multilayered structure, in which the main output neurons subserve the fast and slow electrosensory pathway. The work presented here is focused on the two main output neuron phenotypes of the slow pathway: basilar and non-basilar pyramidal neurons. Their cell bodies are located in the polymorphic layer, and their axons project to the torus semicircularis via the lateral lemniscus (Fig. 1B; Réthelyi and Szabo, 1973; Maler, 1979, Pereira et al., 2014).

The morphology of pyramidal neuron types can be seen in Golgi-stained sections. In both neuron types, basilar (Fig. 1C–H,K) and non-basilar (Fig. 1I,J,L), somata are ovoid or pyramidal and elongated in the dorsal–ventral axis. Basilar pyramidal cells are distinguished by the presence of a single basal dendrite emerging from the ventral region of the soma and extending into the deep neuropil. In some cases, the basal dendrite has three regions: a truncated cone near the soma, often bearing short processes resembling the perisomatic dendrites; an intermediate smooth region of variable length that often bifurcates; and a terminal region with secondary and tertiary varicose dendritic branches (Fig. 1C,D,G). In other cases, the basal dendrite branches as it decreases in diameter and lacks terminal ramifications (Fig. 1H).

To evaluate perisomatic and apical dendritic fields in relation to the orientation of the pisciculi, 13 basilar neurons and six non-basilar neurons were studied in longitudinal Golgi-stained sections whereas eight of each type were studied in transverse sections with respect to the main axis of the ELL.

Perisomatic non-spiny dendrites were thin and short, and sometimes exhibited subterminal branching points (Fig. 1F). The two cell types had similar numbers of perisomatic dendrites extending further in the longitudinal than in the transverse plane (Tables S1 and S2).

Apical dendritic trees are also similar in both neuron types. The number of primary apical dendrites observed in either transverse or longitudinal planes varied from one to five in basilar neurons and from one to six in non-basilar neurons, with no significant differences between the cell types (Tables S3) Primary apical dendrites branched in the ventral molecular layer, giving rise to spiny dendrites extending up to the dorsal molecular layer. An abundance of sessile spines, in addition to pedunculated and thin spines, covering even the smallest distal branches, gave these apical dendrites a warty appearance (Fig. 1E,H–I).

The apical dendritic tree of both types of neurons fanned out rostro-caudally, parallel to the plane of the pisciculi and perpendicular to parallel fibers (Fig. 1H), appeared flattened in the transverse plane. In both basilar and non-basilar neurons, the branching angles of daughter apical dendrites were significantly larger in the longitudinal than in the transverse plane (Fig. 1C–L; Table S4). Consistently, the dendritic trees extended three times further in the longitudinal than in the transverse plane (Fig. 1C–L; Table S5).

This study is based on 19 pyramidal neurons recorded with a median membrane potential of at least −55 mV. Nine of these neurons were labeled and could be anatomically identified as basilar or non-basilar pyramidal neurons, according to their similarity with Golgi stains. The action potentials were classified into two groups, taking into account the trajectory of their transmembrane voltages (Fig. 2A,B), first derivatives (Fig. 2C,D) and phase portraits (Fig. 2E,F).

Regarding the time course of the action potentials, we found that: (1) the median depolarizing overshoot in basilar neurons (+27 mV) was significantly larger than in non-basilar neurons (−0.7 mV) (Fig. 2G); (2) the median inter-peak interval of the first derivatives of the action potentials in basilar neurons (0.52 ms) was twice that observed for non-basilar neurons (0.26 ms; Fig. 2C,H); (3) the depolarizing and repolarizing slopes of the action potential were more symmetrical in non-basilar neurons (compare Fig. 2C with D, and Fig. 2I) and (4) the hyperpolarizing after-potentials were sharper in non-basilar neurons. The median interval between the peak of the spike and the hyperpolarizing after-potential in all basilar neurons (2.14 ms) was longer than that in non-basilar neurons (0.78 ms) (Fig. 2J).

Using a three-parameter criterion considering spike duration (Fig. 2H), spike symmetry (Fig. 2I) and delay of the peak of the after-hyperpolarization (Fig. 2J) together, we could infer the anatomical types of 10 other non-labeled neurons: six were classified as putative non-basilar neurons and four as putative basilar neurons.

Furthermore, in basilar neurons, the after-hyperpolarization was often followed by a depolarizing after-potential peaking about 2–4 ms after the spike peak (Fig. 2K, blue traces, arrow). In four of nine basilar neurons, this depolarizing after-potential gave rise to a smaller, phase-locked second spike also followed by an after-hyperpolarization that further hyperpolarized the membrane (Fig. 2K, black traces). Such second phase-locked spikes were not observed in non-basilar neurons (Fisher’s exact test, P=0.0397).

In the absence of stimulation, we observed that eight out of 10 non-basilar neurons showed a slow rising pre-potential eliciting spike firing at regular intervals. In contrast, basilar neurons were in general silent at rest although three basilar neurons were activated at irregular intervals by fluctuations of the membrane potential corresponding to synaptic activity. Thus, the presence of stereotyped pre-potentials driving repetitive firing at rest was considered a second distinctive functional feature of non-basilar neurons (Fig. 2L, Fisher’s exact test on spontaneous repetitive firing versus neuron type P=0.0476).

In some non-basilar neurons (four out of 10) the basal pacemaker activity was regularly interrupted by long-lasting, quasi-stereotyped hyperpolarizing potentials (Fig. 3A). These hyperpolarizing potentials could also be recruited by extracellular stimulation, confirming their synaptic origin (Fig. 3B). Hyperpolarization of the membrane to −150 mV reduced but did not invert these long-lasting potentials. Similar quasi-periodic, long-lasting inhibitory synaptic potentials were also seen in basilar neurons (Fig. 3D).

Morphology and responses to an increasing series of constant current steps (500 ms duration pulses) are illustrated for typical non-basilar neurons (Fig. 4, n=10) and basilar neurons (Fig. 5, n=7). The electrophysiological profiles of the two phenotypes are compared below.

Spike thresholds were significantly different in non-basilar and basilar neurons, illustrated by the different intersections of the abscissae in Fig. 6 (rank sum test, P=0.001). In non-basilar neurons, spike-firing rate increased monotonically with the intensity of outward current steps, increasing by one spike per second every 3–4 pA, with a mild deceleration at high stimulus intensities (Fig. 6A). Only two non-basilar neurons were not spontaneously active, and these had a very low threshold, close to zero current. In contrast, basilar neurons had a sigmoid response curve, with a higher threshold, ranging from 19 to 210 pA, and a slower increase in spike-firing rate as a function of stimulus step amplitude. When basilar neurons were stimulated at higher intensities, the spike rate became stabilized, reaching a plateau (Fig. 6B) frequency at about 140 Hz; less than half the maximal frequency observed in non-basilar neurons.

After the offset of a depolarizing pulse the activity of the neuron was silenced for a period that increased as a function of the spike rate at the end of the stimulus (Fig. 7F). In addition, in some non-basilar neurons the negative peak of the after-hyperpolarization was followed by a ‘sag’ (Fig. 7G,H, red arrow). Furthermore, in all non-basilar neurons, the offset of hyperpolarization was followed by one or two shorter inter-spike intervals (Fig. 4B, bottom row, and Fig. 7G,H, red spikes). This rebound phenomenon was absent in basilar neurons.

Spike trains started with a delay following depolarizing current onset in basilar neurons, as illustrated in Fig. 8A (median latency=70 ms, n=8). Long latency activation and the differences in spiking probability during a sustained stimulus step were further explored in one neuron by applying 35 successive outward current steps of an intensity 1.5 times the threshold. Excitability increased with time during a sustained depolarization as illustrated by the superimposed responses to 10 of these stimulus steps in Fig. 8B, and the probability of spiking as a function of the latency from the step onset in Fig. 8C. At low intensities, the ratio between the mean intervals recorded during the first and last 250 ms of the stimulus step was significantly less than one, confirming that basilar neuron excitability increases as stimulus is maintained. This suggests the presence of a fast-activating, low-threshold conductance that shunts the membrane after the stimulus onset and inactivates with time.

The amplitude of basilar neuron action potentials decayed with the duration of the stimulus (Figs 5D and 8D). However, at high intensities the ratio between early and late spike rate converged to one, indicating that the effect of spike-firing adaptation on the firing frequency is small.

Four out of nine basilar neurons discharged in doublets (Figs 2K and 5B top row) contrasting with the exclusively single spiking behavior of all non-basilar neurons. Doublet-spiking basilar neurons showed three different types of spiking patterns in response to stimuli of different current intensities. At low stimulus intensities, they fired single large spikes starting at long delays after the stimulus onset. At intermediate stimulus intensities, they fired doublets. At higher stimulus intensities, beyond a crucial point there is a transition in behavior to a constant frequency firing independent of increasing current intensity. This dynamic behavior also evolved with time during a long constant stimulus, as illustrated in Fig. 8D, where at the onset two single spikes (red) are followed by doublet firing (blue) and then, after 400 ms, by a transition to single spiking at a high constant frequency (black).

The responses of two non-basilar neurons and two basilar neurons to stimulation with current ramps were explored (Fig. 9). In non-basilar neurons depolarizing ramps smoothly increased the IF as a monotonic function of current intensity (blue traces in Fig. 9A,B); hyperpolarizing ramps decreased IF to zero (blue traces in Fig. 9C). The neuron of the example showed a nearly periodic activity at rest as indicated by the autocorrelation histogram (Fig. 9D). The magnitude of the change in IF was a function of stimulus intensity (Fig. 9E). As in the case of constant current stimulus, cessation of the depolarizing current caused a pause (red portion of the trace in Fig. 9A,B) while cessation of hyperpolarizing current caused spiking rebound with significantly shorter intervals separating the first two to three spikes (red spikes in Fig. 9C,F). The duration of the post-depolarization pause increased with the amplitude of the ramp and the frequency of the spiking response (Fig. 9G).

Basilar neurons showed a complex dynamic. Ramps of 100 pA s⁻¹ were smoothly followed by the membrane potential up to a critical depolarization value of about 40 mV (Fig. 10A). Above this threshold, the injected current recruited a train of large, low-threshold spikes whose inter-spike intervals then increased monotonically with the instantaneous transmembrane voltage value. Towards the end of a long ramp depolarization (100 pA s⁻¹, 5 s duration), the spikes were followed by a depolarizing after-potential (DAP) of increasing amplitude but this never caused doublets. Ramps of slopes equal to or greater than 200 pA s⁻¹ (Fig. 10B) showed that this depolarizing after-potential progressively increases in amplitude (Fig. 10C). The middle traces in Fig. 10C show spike numbers 1, 33, 67, 100 and 134, illustrating the increasing DAP.

There is a crucial transition point where small spikes begin to arise from the DAP producing an alternating large and small spike-firing pattern (Fig. 10B, blue spikes, magnified in 10C, right inset). As the ramp increased beyond this point the firing pattern changed again, to a nearly constant IF of about 140 Hz, which is then independent of the stimulus intensity and seems to be characteristic of this cell type (see also Fig. 6B).

The plot in Fig. 10D shows that, for 100 pA s⁻¹ ramps, IF increased monotonically as a function of stimulus strength (Fig. 10D, red dots). With steeper ramp increasing at 200 mV s⁻¹, the crucial transition point and behavior shift to a maximum IF was reached at 750 pA. However, as the ramps became yet steeper, although still with the same duration, the crucial transition point (red arrows in Fig. 10D–F) came at higher current intensities, e.g. at around 800 pA for a ramp increasing at 300 mV s⁻¹ and at about 950 pA for a ramp increasing at 400 mV s⁻¹ (Fig. 10D–F). Thus, to reach the transition point leading to a change in firing regime required more current when the ramp depolarization was steeper.

The organization of inputs to the apical dendritic arborization, and to the perisomatic and basilar dendrites in the pulse gymnotiform G. omarorum is similar to that described in wave Gymnotiformes, even though the spatial orientation of the apical dendritic tree and parallel fibers appears to differ (Maler, 1979; Saunders and Bastian, 1984).

The anatomical similarity between these subtypes in wave and pulse Gymnotiformes suggests that basilar neurons are ON cells, excited by primary afferents contacting their basal dendrite (Maler, 1979, 2009a,b). Non-basilar neurons have been designated OFF cells in wave Gymnotiformes, thought to be inhibited by electrosensory input, indirectly via deep granule cells. However, Maler et al. also showed that one type of deep granule cells forms dendro-somatic and dendro-dendritic gap junctions with non-basilar pyramidal neurons thus forming a pathway potentially able to produce indirect, delayed excitation (Maler et al., 1981; Berman and Maler, 1999). This putative, indirect excitatory input, together with our physiological findings showing pacemaker activity, suggest that non-basilar neurons may be both up- and down-modulated by electrosensory input and may thus have additional functions beyond those expected for simple OFF neurons.

Basilar neurons are either silent or show excitatory synaptic activity evoking isolated or paired spikes. In contrast, non-basilar neurons show a basal pacemaker membrane oscillation driving regular spiking activity. The contribution to the pacemaker activity of a hyperpolarization-activated conductance is suggested by the after-hyperpolarization ‘sag’, and by the fast spiking rebound following offset of hyperpolarizing pulses that we observed in most non-basilar neurons. Several other non-exclusive mechanisms may also subserve this type of pacemaker activity, as described, for example, in rat deep cerebellar nuclei, including recruitment of T-type calcium channels, potassium channels and a ‘tunnel effect’ of sodium channels (Dykstra et al., 2016).

A striking synaptic phenomenon observed in both phenotypes is the recurrent emergence of long-lasting hyperpolarizing potentials that periodically interrupt spontaneous firing. These potentials could also be recruited, singly, by extracellular stimulation. As it was not possible to invert these hyperpolarizing potentials, even when the membrane potential at the recording site was hyperpolarized to −150 mV, we infer that these hyperpolarizations correspond to synaptic inhibitory potentials originating electrotonically far from the current injection site. These recurrent long-lasting potentials correspond to those described previously as ‘slow oscillations’ in in vitro preparations in wave Gymnotiformes (Turner and Maler, 1999).

The differences in action potential shape, after-potentials and responsiveness between neuron types allow us to speculate on the underlying conductances shaping the intrinsic properties of each type of neuron. However, we must stress two caveats. First, different conductance combinations may support the same neuronal behavior (Ratliff et al., 2018 preprint; Alonso and Marder, 2020). Second, distinct combinations of appropriately weighted synergistic and antagonistic membrane channel features may result in the same neuronal behavior (Turrigiano et al., 1994).

Action potential shapes and spiking responses provide three arguments consistent with this hypothesis.

First, non-basilar neuron spikes are smaller and sharper than those in basilar neurons. This implies differences in the time constants of the action potential recovery parameters, which are most likely dominated by distinct K⁺ channel repertoires characteristic for each neuron phenotype (Migliore and Shepherd, 2005). Amongst these, one must stress the potential presence of small and big calcium-activated conductances differing in their voltage dependence and time constants. As observed in other systems, small and big Ca²⁺-activated K⁺ conductances may explain the medium negative after-potential (Madison and Nicoll, 1984), and contribute to the sharp repolarization, respectively, characterizing basilar and non-basilar neurons action potential shapes (Adams et al., 1982).

Second, while the spontaneously active non-basilar neurons respond in a phasic–tonic way to depolarizing step stimuli, basilar neurons, which are silent at rest, respond with a slowly increasing firing rate. Non-basilar neurons fire regularly at rest and respond to depolarizing step onsets with an initial marked increase in IF followed by a gradual reduction, following a negative power law, to a maintained lower IF. The offset of depolarization causes a pause followed by a relaxation to the resting frequency. The non-mutually exclusive mechanisms responsible for this type of spike-frequency adaptation include (Benda and Hennig, 2008) M-type (Brown and Adams, 1980) and calcium-activated K⁺ conductances (Madison and Nicoll, 1984), and slow recovery from inactivation of the fast Na⁺ conductance (Fleidervish et al., 1996). Parsimony suggests that the presence of calcium-activated K⁺ conductances should be explored in the search of a common factor determining adaptation in both neuron types and the accelerated repolarization and fast hyperpolarizing after-potentials of non-basilar neurons.

Third, the higher threshold of responses to depolarization and the long delays of the first evoked spikes characteristic of basilar neurons might be explained by a low-threshold-inactivating K⁺ conductance (I_A) shunting membrane resistance and thus reducing excitability after the rapid change in transmembrane voltage at the beginning of a current step. This may contribute to the low-pass filtering properties of the basilar neurons.

Comparative arguments are also consistent with this hypothesis, because in wave fish, ELL output neurons exhibit a wide variety of K⁺ conductances, including small calcium-activated (Ellis et al., 2008) and low threshold conductances, the expression of which is characteristic of neuron type, somatotopic map and species (Fernandez et al., 2005a,b; Mehaffey et al., 2006; Ellis et al., 2008; Mehaffey et al., 2008a).

In half of basilar neurons but in none of non-basilar neurons reported here, the after-hyperpolarization was followed by a depolarizing potential that gave rise to spike doublets. Our results in the pulse fish G. omarorum show that spiking discharges in doublets is a time-dependent phenomenon. Doublet spiking only occurred after rapid transitions in membrane potential, e.g. at the onset of depolarizing current steps or at transition points in fast ramps. They were not present in response to slow depolarizing ramps. This is consistent with the postulated presence of a transiently activated I_A conductance, which prevents basilar neurons from jumping immediately from silence to the high-frequency oscillatory regime observed at the end of long constant current pulses and fast stimulation ramps. A similar process was extensively studied in wave Gymnotiformes (Turner and Maler, 1999; Mehaffey et al., 2008a,b) and described as follows. A spike discharge at the somatic level back-propagates into the apical dendrite. Here, the inward current produced by the dendritic action potential acts as a current source depolarizing the soma, thus producing the depolarizing after-potential. When this current causes a sufficient depolarization, the cell fires a second spike. This activation loop shows characteristic differences among wave species (Mehaffey et al., 2008a). Although this basic loop can be the same in all species, in pulse fish, high-intensity stimuli cause sustained firing at constant frequency instead of the burst exhausting due to dendritic Na⁺ channel inactivation likely present in wave fish (Fernandez et al., 2005a,b). This may suggest species differences in the dendritic repertoire of voltage-dependent conductances.

In the context of the present knowledge of the circuitry and unitary data recorded in acute (Pereira et al., 2014) and freely moving fish (Rodríguez-Cattáneo et al., 2020), our present data suggest different neurocomputational roles for basilar and non-basilar neurons in the ON and OFF output channels of the ELL in pulse species. It is important to recall that field potentials and unitary recordings in the ELL of G. omarorum show four distinct periods of activity in the response to the EOD: an early activation (3–5 ms), followed by a brief silence (5–9 ms), then a mid-term activation (9–20 ms), and lastly a late activation (after 20 ms) (Pereira et al., 2005, 2014).

In OFF units, the post-EOD spiking histogram shows a strong depression starting 4 ms after the EOD followed by a rebounding increase in spike-firing probability in the late phase of the inter-EOD interval (Pereira et al., 2014). We suggest that this self-driven EOD-evoked downregulation in non-basilar neuron pacemaker activity is mediated by inhibitory interneurons, activated by the arrival of the sensory afferent spike train (Berman and Maler, 1998; Maler and Mugnaini, 1994). This inhibitory effect depends on the magnitude of the sensory stimulus, and is followed by an increase in spiking probability towards the end of the EOD cycle and the beginning of the next.

In addition, the observed spike-frequency adaptation and the proposed lateral excitation via gap junctions with granule cells (Maler et al., 1981; Berman and Maler, 1999) may support higher resolution in spatial and temporal contrast, and enable non-basilar neurons to encode transient components of stimuli caused by the movement of nearby textured objects surface facing the electrosensory mosaic (Caputi et al., 2011). This hypothesis is consistent with the paradoxical increase seen in the post-EOD firing rate of OFF units when a textured metal object is moved over the receptive field in a freely moving fish (Rodríguez-Cattáneo et al., 2020).

The electrophysiological behavior of basilar neurons can be related to that of ON units recorded in vivo, for which the post-EOD histogram shows an erratic early spike reflected in a small early peak at about 3–4 ms, followed by the main peak that occurs between 9 and 15 ms after the EOD (Pereira et al., 2014; Rodríguez-Cattáneo et al., 2020) and often showing doublets (fig. 5 in Pereira et al., 2014). The firing pattern may be explained by a direct synaptic depolarizing input caused by the high-frequency burst of 2–5 spikes in primary afferent fibers, arriving between 3 and 12 ms after the EOD (Rodríguez-Cattáneo et al., 2017; Caputi and Aguilera, 2019). However, the silence separating the two peaks of the histogram (4–8 ms after the EOD) is probably caused by feed-forward inhibition originating from deep oval and multipolar neurons that are also excited by primary afferent input and which project to the basilar neuron basal dendrites (Maler et al., 1981; Bastian et al., 1993; Berman and Maler, 1998, 1999). This inhibitory path is also present in G. omarorum (M.C. and A.A.C., unpublished data). This inhibition followed by excitation may provoke a single or a doublet of spikes depending on tonic excitation history and the contextual control made by descending electrosensory paths (Bastian, 1986; Turner and Maler, 1999; Mehaffey et al., 2008b). The low-pass responsiveness of these neurons may facilitate the evaluation of slow changes in the electric image caused by the movements of large or distant objects projecting broad images on the skin (Pereira and Caputi, 2010; Clarke et al., 2014).

In the visual cortex ON and OFF neurons show different receptive fields and response kinetics, suggesting a different role of ON and OFF paths in the evaluation of visual texture and movement (Jansen et al., 2019; Schiller et al., 1986). Similarly, the differences observed in the predominantly tonic and phasic–tonic responsiveness of basilar and non-basilar neurons may provide the fish with the ability of better evaluating a series of electric images corresponding to static or moving objects. This strategy at the early sensory relays is not exclusive to pulse weakly electric fish. Although different or more complex cellular mechanisms may be involved, the presence of tonic and phasic–tonic dual outputs for separately streaming the current input and its dynamic variation in early sensory relays is well known in invertebrates (Reichardt and Poggio, 1976), lower vertebrates (Lettvin et al., 1959) and more recently in the mouse retina (Kim et al., 2015).

Our study shows differences in the intrinsic properties of basilar and non-basilar neurons, which in the context of the known circuitry equip them to perform movement detection in addition to their commonly accepted role in contrast evaluation. We propose that basilar and non-basilar pyramidal neurons, respectively, encode sustained global stimuli, elicited, e.g. by large or distant objects or the EODs of distant fish, and transient textural stimuli, such as those elicited by sharp edges of moving nearby objects or the complex near field of a neighboring conspecific.

The authors thank Dr Kirsty Grant for her detailed critical evaluation of the manuscript, her insightful comments and contributions on both science and language style, and the final English edition of the submitted manuscript. The authors also thank Drs Curtis Bell and Victor Han for their help with setting up the whole-cell patch recording technique, and the technical assistance of Mag. María Inés Rehermann and Sofía Ramundey during the preparation and quantification of Golgi-stained material. This work is part of the doctoral thesis studies of J.N. (Nogueira Borde, 2011).