Neuronal mechanisms for object discrimination in the weakly electric fish Eigenmannia virescens

Many species of South American gymnotoid and African mormyriform fishes actively emit weak electrical potentials through their electric organs. These fish and others possess sensitive electroreceptors in the skin which enable them to detect minute electrical potentials in the surrounding water. The electric organ discharge (EOD) in these weakly electric fish is shown to be involved in social communication, reproductive and agonistic interactions as well as in object location and body orientation (Lissmann & Machin, 1958; Bullock, 1968a; Belbenoit, 1969, 1970; Black-Cleworth, 1970; Valone, 1970; Bullock, Hamstra & Scheich, 1972; Hopkins, 1972a, b, 1974; Harder, 1972; Heiligenberg, 1973, 1974; Bell, Myers & Russell, 1974; Russell, Myers & Bell, 1974; Westby, 1975; Kramer, 1974; Meyer, Bullock & Heiligenberg, 1976).

The electric organ discharge of an individual fish produces a voltage gradient along its longitudinal body axis, forming an approximately dipole electric field. The form of this field has been deduced experimentally (Knudsen, 1975) and simulated on a computer (Heiligenberg, 1975). In general the field varies with body geometry, particularly with the length of a fish as well as with the position of its tail. The electric field becomes distorted when an object with different electrical properties from the surrounding water is introduced near the fish. Electrolocation in its active mode is based on the fish’s ability to detect local distortion of its electric field due to the presence of the object (Lissman & Machin, 1958). Since objects with varying conductivities distort the field differently, the fish can conceivably discriminate the conductivity of an object. A number of electrophysiological studies have revealed that information about the conductivity of objects in the fish’s field is encoded in the primary afferent fibres of the lateral line nerve (Hagiwara, Kusano & Negishi, 1962; Hagiwara, Szabo & Enger, 1965; Hagiwara & Morita, 1963; Scheich, Bullock & Hamstra, 1973; Schlegel, 1975), lateral line lobe of the medulla (Enger & Szabo, 1965; Schlegel, 1973, 1974) and the cerebellum (Bastian, 1974, 1975) by any of several encoding mechanisms (Bullock, 1968 b).

We know, however, that conductivity alone does not completely characterize the electrical properties of some objects. Since many living tissues have a capacitive element (Cole, 1942; Heiligenberg, 1973), it could be that part of the electric fish’s ability to discriminate objects can be attributed to detection of the capacity of an object. The experiments described here were designed to test the hypothesis that weakly electric fish of the genus Eigenmannia possess physiological mechanisms capable of encoding the resistive and capacitive properties of objects.

Scheich et al. (1973) found in preliminary experiments that some afferent fibres in the lateral line nerve of Eigenmannia can respond in a different fashion to a purely resistive shunt between two electrodes a few centimetres apart in the water a few centimetres from the fish, than they do to a capacitive shunt of the same impedance at the fish’s EOD frequency. Other afferent fibres did not discriminate between but did respond to both capacitive and resistive shunts with equal sensitivity when the impedance was given at EOD frequency. It seemed in their small sample that the class of afferents called phase coders or T units (Bullock & Chichibu, 1965 ; Scheich et al. 1973 ; Hopkins, 1976) was discriminating whereas so-called probability coders or P units were equally sensitive to both. However, the number of units studied was small and we will show in this report that both classes of units are able to encode local resistive and capacitive shunts differentially.

Experimental animals were 18 Eigenmannia virescens, South American weakly electric fish of the superfamily Gymnotoidei. They ranged in length from 9 cm to 16 cm. They were kept in water of 2 kii cm resistivity (obtained by mixing deionized water with aquarium salt formulated by Huckstedt) in 801 tanks at 24 °C. Two different immobilized preparations were made, one by injecting fish with 0-02 mg of gallamine triethiodide (Flaxedil) to make an electrically silenced preparation, the other by anaesthetizing fish with tricaine methanesulphonate (MS222), at 65 parts per million, to make an immobilized preparation with unsilenced EOD. Flaxedil silences the electric organ of Eigenmannia which is myogenic (Bennett, 1971). MS222 only reduces the EOD amplitude slightly with minimal effect on respiratory movements.

With electrically silenced fish, a sine wave stimulus, simulating the fish’s EOD, was presented to the fish via two carbon rods (30 cm apart) at extension of the fish’s longitudinal body axis. The sinusoidal stimulus allowed a precise measure of the shunting impedance. It also facilitated classification, and measurement of threshold of response and best excitatory frequency, of units of the lateral line nerve from which recordings were made. Furthermore, since the sinusoidal stimulus intensity was adjustable, a unit’s response to a local impedance shunt could be investigated at various stimulus intensities. Note, however, that the electric field generated by these dipole electrodes differed considerably from the self-generated electric field because the fish’s electric organs do not represent a simple dipole source (Heiligenberg, 1975). The results obtained under the simulated field must therefore be compared with those obtained under the self-generated field.

A small part of the fish’s body (1 cm) just behind the gill was wrapped in a plastic screen which was attached to a Plexiglas holder above the water surface. Most of the fish’s body except the head was immersed in the water. The experimental tank consisted of a Plexiglas container with a dimension of 25 × 40 × 9 cm. The fish was artificially respired with aerated water through its mouth and gills. The respiration water and the bath water were at room temperature (22 °C) and resistivity was maintained at 2 kΩ cm or occasionally at 20 kΩ cm.

The posterior branch of the anterior lateral line nerve was exposed just above the gill. Single unit recordings were made with 3M-KCl-filled micropipettes (impedance = 20–50 MΩ) from the intact nerve or with hooked tungsten electrodes from the distal end of the cut nerve (the nerve was sectioned just before it entered the skull in this case). Recordings were made differentially with the reference electrode placed immediately adjacent to the active electrode. Action potentials were amplified in a conventional way and monitored audiovisually. Unit response was studied on line with a Nicolet computer (model 1072) and an electronic counter.

In order to observe the differential effects of local capacitance and resistance change on the neuronal response, the receptive field of an individual unit was first identified by local electrical stimulation. The water impedance adjacent (1-10 cm away from the fish’s skin) to the receptor was then modified by shunting capacitors or resistors of chosen values across the bath via two carbon rod electrodes. The distance between these two carbon rods was usually maintained at 2 cm but occasionally also at 4 cm. As measured at the rods, in 2 kΩ cm water, the resistance of water plus rods was 170 Ω and the capacitance was 17 nF. The rods could be rotated and moved to any desired angle and position along the fish’s body. The effect of an impedance shunt on the field potentials was monitored with a pair of wire electrodes (1 cm apart) intermediate between the two carbon rods. The response of an individual cell was studied under various shunting conditions and with different orientations and positions of the shunting electrodes.

A total of 80 units were studied in 18 fish. Of these, 49 units were in electrically silenced fish and 31 units were in anaesthetized fish with self-generated EOD. The results from the two preparations were similar except for some details which we shall describe later.

In Eigenmarmia, there are two major types of primary electroreceptive fibres sensitive to high-frequency sinusoidal electrical stimuli (Scheich et al. 1973). In the curarized fish, these two types of fibres are readily distinguishable. The probability coders or P units were generally spontaneously active in the absence of any applied electrical stimulus. This spontaneous activity was modified in the presence of a sinusoidal electrical stimulus of appropriate amplitude and frequency. Response thresholds ranged from 50 μV/cm to 2 mV/cm. This type of fibre generally required high stimulus intensity, i.e. from 1·8 mV/cm to over 6 mV/cm, in order to obtain 1:1 firing to each stimulus cycle. Furthermore, the phase of spike occurrence was poorly time locked to the stimulus cycle (jitter over 500 μs). The other type of fibre, the phase coder or T unit was only weakly active or completely inactive in the absence of electrical stimuli. These fibres had low response thresholds in comparison with the probability coders as well as lower overall 1:1 firing thresholds (0·3 mV/cm to 1·5 mV/cm). The spike was generally precisely time locked to the stimulus cycle with jitter of less than 100 μs. In each fish, the best excitatory frequencies for both types of fibres were right around or slightly lower than the fish’s EOD frequency as monitored prior to Flaxedil injection (see also Scheich & Bullock, 1974 ; Hopkins, 1976).

The firing patterns of the two fibre types differed considerably in response to a local impedance shunt. The fibre types also responded differently to resistive capacitive shunting impedances and this depended on the applied sinusoidal stimulus intensity as well as on the location and orientation of the shunting electrodes. Unless specified otherwise, all fibres were stimulated with sinusoidal electrical stimuli at their best excitatory frequencies and at intensities of 1, 1·5 or 1·75 mV/cm. For phase coders, additional observation was made at 3 dB above the 1:1 firing threshold of each unit.

At a stimulus intensity of 1–1·75 mV/cm, most probability coders showed firing probabilities of from 0-3 to 0-65. A local impedance shunt either increased or decreased the firing probability of the unit depending on the position of the shunting electrodes relative to the fish. When both shunting electrodes were anterior to the electroreceptor with the electrodes 2 cm apart on a line parallel to the fish’s surface and a lateral distance of 1–3 cm away, an impedance shunt reduced the firing rate. When both shunting electrodes were posterior to or immediately opposite the receptor, an impedance shunt increased the firing rate. A low-impedance shunt was more effective than a high-impedance shunt in altering the unit’s firing frequency. Also, the change in the unit’s firing frequency varied with the element of the impedance. As a rule, the response curve as a function of impedance for a resistance was less steep than that for a capacitance, when plotted as in the typical example in Fig. 1. A table of capacitive reactance is shown in the Appendix for a sinusoidal stimulus frequency of 300 Hz. Unit 8L-2 in Fig. i was best excited by a 300 Hz sinusoidal electrical stimulus. At a stimulus intensity of 1·75 mV/cm the unit fired with a frequency of 183 ± 6 impulses/second (i.p.s.) which was equivalent to a firing probability of 0·61. This firing frequency was altered in the manner just described.

The effect of an impedance shunt seemed to depend on the sinusoidal stimulus intensity level, as illustrated in Fig. 2. At 1 mV/cm, the effect of a capacitance shunt and of a resistance shunt differed only slightly (Fig. 2d) but at 1-5 mV/cm this difference was greatly amplified (Fig. 2b). Thus discrimination between the resistance and the capacitance of an object is sharper at higher intensities. However, when the unit’s firing rate reaches the 1:1 plateau at still higher intensities, such discrimination again deteriorates.

We have shown earlier that the effect of an impedance shunt was dependent on the longitudinal position of the shunting electrodes with relation to the fish (see Fig. 1). We further found that the orientational angle of the shunting electrodes as well as their distances from the fish were also important, as shown in Fig. 3. An impedance shunt was most effective when the shunting electrodes were near the fish (Fig. 3a). The response change to the impedance shunt diminished rapidly when shunting electrodes were moved away from the fish. The orientation of the shunting electrodes was also important, with maximal effect when they were oriented parallel to the fish’s longitudinal body axis (Fig. 3b). Such an effect was probably location specific but this was not explored further.

Most of the phase coders showed 1:1 firing to each stimulus cycle at intensities above 1 mV/cm. An impedance shunt seldom altered the probability of firing of these units except at low stimulus intensities. However, the phase of spike occurrence changed systematically with the impedance except for a few units with receptive fields at the caudal portion of the trunk, in which the phase of firing occurrence was never altered. A low-impedance shunt affects the phase of firing more than a high-impedance shunt. Furthermore, in comparison with the resistance shunt, the capacitance shunt of equivalent impedance was more effective in eliciting the phase shift, as shown in Fig. 4. This is due in part to the fact that capacitance shunts not only affect the amplitude of the stimulus at the receptor but also the phase. However, such phase alteration, being maximal at lower impedances, is not entirely responsible for the discrepancy in the observed neural phase shift as a result of capacitance and resistance shunts as evidenced in Fig. 5. It can be seen that in Fig. 5 the phase or latency differences to the two shunting conditions is not maximal at the lowest impedance. The above assertion is also supported by the fact that direct measurement of the stimulus phase shift at the receptor as a result of capacitance shunt is not as large as the differences in neural latencies observed in the neural data at various impedances. Notice that in Fig. 4 the firing phase shifted later or earlier in the stimulus cycle depending on whether the shunting electrodes were anterior (circles) or posterior to the receptor (squares) when they are 2 cm apart on a line parallel to the fish’s body at a lateral distance of from 1–3 cm.

The differential change in latency was a function of the input stimulus intensity level, being greater at medium intensities than at lower or higher intensity levels, as shown in Fig. 5. It seems that the operating point is of great importance in determining the unit’s response properties to impedance shunt. Notice that in Fig. 5 the water resistivity was 20 kΩ cm instead of 2 kΩ cm and the effect of shunting extended to higher impedances (cf. Fig. 4).

The change in latency or phase shift was also a function of the distance of the shunting electrodes from the fish as well as of the relative orientation of the shunting electrodes. Figure 6 exhibits the response patterns of the phase shift from a unit under the influence of the above factors. Again we can see that the phase shift was maximal when the shunting electrodes were near the fish (Fig. 6a) and when they were parallel to the fish’s longitudinal body axis (Fig. 6b).

In addition to the probability and phase coders in the posterior branch of the anterior lateral line nerve, there was a third type of fibre which was found to be insensitive to electrical stimuli in the range employed (up to 10 mV/cm). These units fired at a high rate, frequently in pairs or triplets, in the absence or presence of applied electrical stimuli of less than 10 mV/cm. In the latter case, the spike occurrence was not time locked to the stimulus cycle. However, some of these units were responsive to externally forced bending of the tail. Bending the tail to the ipsilateral side of the recorded nerve reduced the firing frequency of the unit, whereas bending to the contralateral side increased the firing frequency. This phenomenon was independent of the presence of sinusoidal electrical stimuli. These units probably play an important role in electrolocation and object discrimination, as we shall point out later. It must be noted however, that not all units in this category exhibited this type of response to bending of the tail.

In the anaesthetized fish where the EOD persisted, unit classification was based on the unit’s probability of firing and the precision of its time locking activity (Bullock & Chichibu, 1965). Units that fired 1:1 to the fish’s EOD were classified as phase coders and those that did not as probability coders. In such a preparation, each receptor was being driven by the fish’s own EOD, and therefore at around the unit’s best excitatory frequency. Of course the intensity functions could not be determined. The responses to a local impedance shunt were qualitatively the same as those observed in electrically silenced fish with a simulated EOD.

In response to a local impedance shunt, phase coders altered their phase of firing while maintaining 1:1 firing whereas probability coders varied their firing probabilities. Both probability coders and some phase coders were sensitive to local impedance shunts. Capacitance shunts generally produced a greater change in the response of units than did the resistance shunts, as shown in Fig. 7. The response patterns were not significantly different from those obtained in electrically silenced fish. However, the position of the shunting electrodes along the fish’s body gave varying results in different units. Also the effect of the angle of orientation of these electrodes, being dependent on their anterior-posterior position relative to the fish, was complex. Most commonly, the effect of an impedance shunt was maximal when the shunting electrodes were posterior and near the receptor as shown in Fig. 8. Note, however, that some phase coders maintained their phase of firing regardless of the shunt, the type of shunt or the position of the shunting electrodes. These units probably play a significant role as a time reference for decoding the latency shifts.

Our results demonstrate that Eigenmannia virescens possesses physiological mechanisms for object discrimination on the basis of both capacity and resistance. Both the probability and phase coders are sensitive to a local impedance shunt and both types of fibres show different responses to resistive and capacitance shunts of equivalent impedance. There is as yet no behavioural evidence for this ability and so an obvious opportunity for significant experimental ethology exists. But we have reasons to believe that these physiological mechanisms are behaviourally significant. These fish are nocturnal and often live in environments of poor visibility. Electroreception plays an important role in object location. In the aquarium, normal and blinded fish generally hover around plants, leaves or roots. Leaves of aquatic plants have specific capacitances and resistances in the order of 50 nF/cm² and 100 kΩ cm (Heiligenberg, 1973). We have shown that in 20 kΩ cm water the fish are sensitive to shunting resistances of over 20 kΩ and capacitances of less than 50 nF in the arbitrary geometry tested. This sensitivity is considerably reduced in 2 kΩ cm water. However, the field distortion due to an external impedance shunt is conceivably less than that due to the actual presence of an object of similar impedance in the water at the same location. Therefore the sensitivities of the fibres that we observed may be realistically even higher. Eigenmannia’s natural environments vary widely with reported resistivities ranging from 10 to 150 kΩ cm (Hopkins, 1972; W. Heiligenberg, personal communication). The appearance of a novel object with different impedances and resistances of minute amounts can conceivably be detected. In a few experiments where the fish was anaesthetized, we presented a 2·5 cm live guppy (Lebistes retiadatus) in the fish’s electric field at a distance of 3 cm laterally. The P units being recorded responded with a change of firing rate of over 10 i.p.s. and the Y units with a phase shift of over 40 μs which was equivalent to the response to 68 nF or 10 kΩ. Thus, unit sensitivity is probably behaviourally significant.

The physical basis of these differences in response to capacitance and resistance shunts may not be intuitively obvious. For a sine wave in a simple circuit with no parallel paths, C orR of the same impedance cause the same voltage drop ; the only difference is the phase. In the three dimensional case, in water there is always a low R_water in parallel with the C_shunt or R_shunt introduced across a pair of electrodes. Current through the water, to the extent that it is purely resistive, is not phase shifted. The C_shunt and R_shunt mix phase-shifted and phase-unshifted current. The voltage drop measured at any point away from the shunting electrodes is therefore different for the C_shunt and R_shunt of equal impedance. Receptors in various positions will see unequal values of this difference. The amplitude change may be small and yet significant because of the high increment sensitivity of tuberous receptors (Fig. 5a inset).

For EOD signals which have higher harmonic components, the wave form of the EOD signal is likely to be different at the receptor for the two shunting conditions due to differential phase shift by the various spectral components of the EOD. The experiments reported indicate that under our conditions there is no significant difference between the responses in the presence of the EOD and those with a pure sine wave at the EOD fundamental. This is understandable on the basis of the relatively low harmonic content (Bullock, Behrend & Heiligenberg, 1975) and the fall in sensitivity of receptors above the EOD fundamental (Scheich et al. 1973; Hopkins, 1976).

The shifts reported in the time of firing of T units must be the mixed result of the phase shift of the field and the ‘phase coding’ of ampEtude change by the receptor. The latter is larger in our geometry. A C_shunt at the high end of the curves presented (20 μF) caused a maximally 60 μs shift measured near the fish, whereas a T unit could shift 250 μs.

The usefulness of active electroreception for object detection has been thought to be limited by a severe effect of distance. The EOD field strength has been thought to fall off with the 3rd power of distance from the fish as it should for a small dipole in the far field. The influence of an object, representing an impedance discontinuity that can theoretically be treated as a source, has also been thought to fall off as the 3^rd power of the distance. Together, the effect of distance on object detection has been supposed to be a 6th power function. Heiligenberg (1975) found in his computer simulation that the fall off is much less severe, due to the relevant distances being not very different from the size of the fish and the shape of the EOD field being not that of the simple theoretical source. Our measurements of actual receptor responses also show a much less steep fall with distance. Figures 3, 6 and 9 show the 6th power slope together with actual responses. Most of the units follow an inverse 2nd power function of the distance in agreement with the results from computer simulation (Heiligenberg, 1975).

A comparison of our results with those of Scheich et al. (1973) indicates that there is a discrepancy in the sensitivities of the most sensitive primary electroreceptive neurones to local impedance shunts. They reported a response sensitivity to shunts of several hundreds of kilohms which is one order higher than that reported here. In the light of normal variation among units this is not a great discrepancy. The origin of this discrepancy is not entirely clear but it probably resides partly in the different techniques employed for holding the fish during the recording sessions. In the present study, the fish was attached to a Plexiglas holder on its dorsal surface with a small plastic screen so that the electric field was practically unchanged. Scheich et al. held the fish by attaching one side on a wax mould which might alter the electric field somewhat more extensively. Furthermore Scheich et al. used anaesthetized fish with a natural EOD and a slightly different arrangement of shunting electrodes.

The results from electrically silenced fish indicate that in a simulated dipole field the fish can increase its discriminating ability by adjusting the electrical potential at the receptor to appropriate levels. We have shown that the differential responses to resistance and capacitance shunts are optimal at stimulus intensities ranging from 1-5 to 2·5 mV/cm varying from one fish to another, and with fibre type. A fish can modify local electrical potentials by moving its tail laterally, as shown on a computer simulation by Heiligenberg (1975). This action is frequently observed when a novel object is introduced into the vicinity of an Eigenmannia. It has been proposed that such action maximizes the electrical ‘shadow* of the object on the fish’s body wall (Bastian, 1974.; Heiligenberg, 1975). In view of the present results, we may add that lateral tail movement probably also enhances the fish’s discriminatory ability. The fish additionally possesses information about the position of its tail. A number of fibres in the posterior branch of the anterior lateral line nerve are sensitive to the direction and the degree of forced bending of the tail. Many lateral line lobe units (T. A. Viancour, personal communication) and cerebellar units (Bastian, 1974) have similar response characteristics. However, the terminal receptors and the origin of this sensitivity are unknown. In mormyrids a rich sensory innervation in the narrow tail peduncle, which experiences the maximum flexion with tail movements has been described (Bruns, 1971) but this has not been looked for in gymnotids to our knowledge.

In addition to the electrical properties of an object, the electric fish primary electroreceptors are also sensitive to orientation and distance of the object. In a single unit the change in firing rate or in phase is therefore ambiguous as to whether the cause is a change in the resistive load or a complex impedance change, or even in position, or orientation, distance or size of an inhomogeneity in the water. However, the large number of receptors and large central structures devoted to this modality suggest that parallel channel coding and decoding may permit the fish to sort out these parameters, especially if the fish is allowed to move and integrate over time. A number of studies have shown that lateral line lobe and cerebellum are intimately involved in object location. Bastian (1974) has also found that some cerebellar units are differentially responsive to capacitance and resistance shunts but we know nothing yet about the properties of lateral line lobe units. This is clearly an area for future research.

Besides the South American wave species such as Eigenmannia, many electric fish of the family Mormyridae in Africa have a pulse type of EOD that is extremely brief (0-3 ms) with considerable energy above 3 kHz. These fish should be even more sensitive to small capacitances. We have not yet examined their receptors.

It seems that electrolocation and object discrimination in the weakly electric fish and the ability of bats to echolocate and discriminate objects (Griffin, Friend & Webster 1965; Webster, 1967; Suthers, 1965) share a common basis. They both rely upon the ability of the organism to detect and analyse the pattern of the returning signal which they initially emit. In bats, it has been shown that the ability of target discrimination in some species relies on the availability of frequency dependent cues, while in other species it does not (Bradbury, 1970; Simmons et al. 1974). In the electric fish, the cues for object discrimination consist of frequency-dependent (capacity) and frequency-independent (resistivity) cues but it remains to be shown behaviourally that the electric fish actually utilizes these cues.

Capacitive reactances on the basis of a sinusoidal stimulus frequency of 300 Hz

We thank K. Behrend, K. S. Cole, W. Heiligenberg, H. Krausz, D. Lange, T. G. Uter and T. A. Viancour for the vigorous discussions throughout the study, F. W. Bloodworth, W. P. Reetz and T. G. Uter for their technical assistance, and Lynne Sasse for her expert typing. This study is supported by the National Institutes of Health and National Science Foundation grants to T. H. Bullock.