Abstract
Sensory systems evolve in the ecological niches that each species is occupying. Accordingly, encoding of natural stimuli by sensory neurons is expected to be adapted to the statistics of these stimuli. For a direct quantification of sensory scenes, we tracked natural communication behavior of male and female weakly electric fish, Apteronotus rostratus, in their Neotropical rainforest habitat with high spatiotemporal resolution over several days. In the context of courtship, we observed large quantities of electrocommunication signals. Echo responses, acknowledgment signals, and their synchronizing role in spawning demonstrated the behavioral relevance of these signals. In both courtship and aggressive contexts, we observed robust behavioral responses in stimulus regimes that have so far been neglected in electrophysiological studies of this well characterized sensory system and that are well beyond the range of known best frequency and amplitude tuning of the electroreceptor afferents' firing rate modulation. Our results emphasize the importance of quantifying sensory scenes derived from freely behaving animals in their natural habitats for understanding the function and evolution of neural systems.
SIGNIFICANCE STATEMENT The processing mechanisms of sensory systems have evolved in the context of the natural lives of organisms. To understand the functioning of sensory systems therefore requires probing them in the stimulus regimes in which they evolved. We took advantage of the continuously generated electric fields of weakly electric fish to explore electrosensory stimulus statistics in their natural Neotropical habitat. Unexpectedly, many of the electrocommunication signals recorded during courtship, spawning, and aggression had much smaller amplitudes or higher frequencies than stimuli used so far in neurophysiological characterizations of the electrosensory system. Our results demonstrate that quantifying sensory scenes derived from freely behaving animals in their natural habitats is essential to avoid biases in the choice of stimuli used to probe brain function.
Introduction
Sensory systems evolve in the context of species-specific natural sensory scenes (Lewicki et al., 2014). Therefore, naturalistic stimuli have been crucial for advances in understanding the design and function of neural circuits in sensory systems, in particular the visual system (Laughlin, 1981; Olshausen and Field, 1996; Gollisch and Meister, 2010; Froudarakis et al., 2014) and the auditory system (Theunissen et al., 2000; Smith and Lewicki, 2006; Clemens and Ronacher, 2013). Communication signals are natural stimuli that are, by definition, behaviorally relevant (Wilson, 1975; Endler, 1993). Not surprisingly, certain acoustic communication signals, for example, have been reported to evoke responses in peripheral auditory neurons that are highly informative about these stimuli (Rieke et al., 1995; Machens et al., 2005). However, other stimuli that do not strongly drive sensory neurons may also be behaviorally relevant and equally important for understanding the functioning of neural systems. Unfortunately, they are often neglected in electrophysiological studies because they do not evoke obvious neural responses (Olshausen and Field, 2005).
To address this bias, we quantified behaviorally relevant sensory scenes that we recorded in freely interacting animals in their natural habitat. Tracking the sensory input of freely behaving and unrestrained animals in natural environments is notoriously challenging (Egnor and Branson, 2016). We took advantage of the continuously generated electric organ discharge (EOD) (Fig. 1A) of gymnotiform weakly electric fish to track their movements and electrocommunication signals without the need for tagging individual fish.
The quasi-sinusoidal EOD, together with an array of electroreceptors distributed over the fish's skin (Carr et al., 1982), forms an active electrosensory system used for prey capture (Nelson and MacIver, 1999), navigation (Fotowat et al., 2013), and communication (Smith, 2013). Both the EOD alone and its modulations function as communication signals that convey information about species, sex, status, and intent of individuals (Hagedorn and Heiligenberg, 1985; Stamper et al., 2010; Fugère et al., 2011). In Apteronotus, several types of brief EOD frequency (EODf) excursions called “chirps” (Fig. 1B) have been studied extensively in the laboratory setting (Engler and Zupanc, 2001) and have been associated with courtship (Hagedorn and Heiligenberg, 1985), aggression (Zakon et al., 2002), and the deterrence of attacks (Hupé and Lewis, 2008). P-unit tuberous electroreceptors encode amplitude modulations of the EOD (Bastian, 1981a) as they are induced by the presence of a second fish and by chirps (Benda et al., 2005; Walz et al., 2014).
Here, we describe electrocommunication behavior of weakly electric fish recorded in their natural Neotropical habitat with unprecedented high temporal and spatial resolution. We found extensive chirping interactions on timescales ranging from tens of milliseconds to minutes in the context of courtship. In a complementary breeding experiment, we confirmed the synchronizing role of chirping in spawning. From the observed courtship and aggression scenes, we computed the statistics of interaction distances determining the effective signal amplitudes and the signal frequencies driving the electrosensory system. In the discussion, we compare these natural stimulus statistics with the known coding properties of electroreceptor afferents.
Materials and Methods
Field site
The field site is located in the Tuira River basin, Province of Darién, Republic of Panamá (Fig. 1-1A), at Quebrada La Hoya, a narrow and slow-flowing creek supplying the Chucunaque River. Data were recorded ∼2 km from the Emberá community of Peña Bijagual and ∼5 km upstream of the stream's mouth (8° 15′ 13.50” N, 77° 42′ 49.40” W). At our recording site (Fig. 1-1B), the water level ranged from 20 cm at the slip-off slope to 70 cm at the cut bank. The water temperature varied between 25°C to 27°C on a daily basis and water conductivity was stable at 150–160 μS/cm. At this field site, we recorded four species of weakly electric fish, the pulse-type fish Brachyhypopomus occidentalis (∼30–100 Hz pulses/s), the wave-type species Sternopygus dariensis (EODf at ∼40–220 Hz), Eigenmannia humboldtii (200–580 Hz), and A. rostratus (580–1100 Hz). We here focused exclusively on A. rostratus, a member of the A. leptorhynchus species group (brown ghost knifefish; de Santana and Vari, 2013).
Field monitoring system
Our recording system (Fig. 1C and Fig. 1-1B) consisted of a custom-built 64-channel electrode and amplifier system (NPI Electronics) running on 12 V car batteries. Electrodes were low-noise head stages encased in epoxy resin (1× gain, 10 × 5 × 5 mm). Signals detected by the head stages were fed into the main amplifier (100× gain, first-order high-pass filter 100 Hz, low-pass filter 10 kHz) and digitized with 20 kHz per channel with 16-bit amplitude resolution using a custom-built low-power-consumption computer with two digital-analog converter cards (PCI-6259; National Instruments). Recordings were controlled with custom software written in C++ (https://github.com/bendalab/fishgrid) that also saved data to hard disk for offline analysis (exceeding 400 GB of uncompressed data per day). We used a minimum of 54 electrodes arranged in a 9 × 6 array covering an area of 240 × 150 cm (30 cm spacing). The electrodes were mounted on a rigid frame (thermoplast 4 × 4 cm profiles, 60% polyamid, 40% fiberglass; Technoform Kunststoffprofile), which was submerged into the stream at the cut bank side and fixed in height 30 cm below the water level.
Data analysis
All data analysis was performed in Python 2.7 (www.python.org, https://www.scipy.org/). Scripts and raw data (Panamá field data: 2.0 TB, Berlin breeding experiment: 3.7 TB of EOD recordings and 11.4 TB video files) are available on request, data of the extracted EOD frequencies, position estimates, and chirps are available at https://doi.org/10.12751/g-node.87584d and some of the core algorithms are accessible at Github under the GNU general public license (https://github.com/bendalab/thunderfish).
Summary data are expressed as means ± SD unless indicated otherwise.
Spectrograms in Figures 3 and 7B were calculated from data sampled at 20 kHz in windows of 1024 and 2048 data points, respectively, and shifted by 50 data points.
Fish identification and tracking.
First, information about electric fish presence, EODf, and approximate position were extracted. Each electrode signal was analyzed separately in sequential overlapping windows (1.22 s width, 85% overlap). For each window, the power spectral density was calculated (8192 FFT data points, five subwindows, 50% overlap) and spectral peaks above a given threshold were detected. Individual fish were extracted from the list of peak frequencies based on the harmonic structure of wave-type EODs. Finally, fish detections in successive time windows were matched, combined, and stored for further analysis.
Based on EODf, we separated male (EODf > 750 Hz) from female fish (EODf < 750 Hz) (Meyer et al., 1987). The data allowed us to analyze courtship and aggression of six male and two female fish in detail.
Position estimation.
For each fish, the signals of all electrodes were band-pass filtered (forward–backward Butterworth filter, third order, 5× multipass, ±7 Hz width) at the fish's EODf. Then, the envelope was computed from the resulting filtered signal using a root-mean-square filter (10 EOD cycles width). Each 40 ms the fish position x⃗ was estimated from the four electrodes i with the largest envelope amplitudes Ai at position e⃗i as a weighted spatial average (see Movie 1) as follows: This estimate proved to be the most robust against fish moving close to the edges of the electrode array, as verified with both experiments and simulations (Henninger, 2015). In short, we measured the spatial distribution of an electric fish's EOD field in a large tank (3.5 × 7.5 × 1.5 m, w × l × h) under conditions similar to field conditions (water depth 60 cm, fish and electrode array submerged 30 cm below surface). We used this dataset for evaluating the performance of three algorithms for position estimation and for fitting a simple dipole model for the spatial electric field distribution. The dipole model was then used to evaluate the algorithms in greater detail by simulating stationary and moving fish for various electrode configurations. For the electrode configuration used, the weighted spatial average yielded a precision of 4.2 ± 2.6 cm on level with the electrode array and 6.2 ± 3.8 cm at a vertical distance of 15 cm as computed by extensive simulations. Finally, the position estimates were filtered with a running average filter of 200 ms width to yield a smoother trace of movements.
Chirp detection and analysis.
For each fish, the electrode voltage traces were bandpass-filtered (forward–backward Butterworth filter, third order, 5× multipass, ±7 Hz width) at the fish's EODf and at 10 Hz above the EODf. For each passband, the signal envelope was estimated using a root-mean-square filter over 10 EOD cycles. Rapid positive EODf excursions cause the signal envelope at the fish's baseline frequency to drop and in the passband above the fish's EODf to increase in synchrony with the frequency excursion. If events were detected synchronously in both passbands on more than two electrodes, and exceeded a preset amplitude threshold, they were accepted as communication signals.
Communication signals with a single peak in the upper passband were detected as small chirps. Signals of up to 600 ms duration and two peaks in the upper passband, marking the beginning and the end of the longer frequency modulation, were detected as long chirps. All chirps in this study were verified manually. However, it is likely that some chirps were missed, since detection thresholds were set such that the number of false positives was very low. In addition, abrupt frequency rises (Engler and Zupanc, 2001) were probably not detected because of their low-frequency increase.
Interchirp interval probability densities were generated for pairs of fish and only for the time period in which both fish were producing chirps. Kernel density histograms of interchirp intervals (Fig. 5-1) were computed with a Gaussian kernel with an SD of 20 ms.
Rates of small chirps before and after female long chirps (see Fig. 5A,C) were calculated by convolving the chirp times with a Gaussian kernel (σ = 0.5 s) separately for each episode and subsequently calculating the means and SDs.
For quantifying the echo response (see Fig. 6), we computed the cross-correlogram as follows: with the na chirp times ta,j of fish a and the nb chirp times tb,i of fish b using a Gaussian kernel g(t) with a SD of 20 ms. To estimate its confidence intervals, we repeatedly resampled the original dataset (2000 times jackknife bootstrapping; random sampling with replacement), calculated the cross-correlogram as described above, and determined the 2.5% and 97.5% percentiles. To create the cross-correlograms of independent chirps, we repeatedly (2000 times) calculated the cross-correlograms on chirps jittered in time by adding a random number drawn from a Gaussian distribution with a SD of 500 ms and determined the mean and the 2.5% and 97.5% percentiles. Deviations of the observed cross-correlogram beyond the confidence interval of the cross-correlogram of jittered chirp times are significant on a 5% level and are indicative of an echo response. Reasonable numbers of chirps for computing meaningful cross-correlograms (more than several hundreds of chirps) were available in five pairs of fish.
Beat frequencies and spatial distances.
The distance between two fish at the time of each chirp (Fig. 8B) was determined from the estimated fish positions. As the receiver of the chirp, we assigned the fish that was closest to the sender and at maximum 150 cm away. The distance estimates were compiled into kernel density histograms that were normalized to their maximal value. The Gaussian kernel had a SD of 1 cm for courtship small chirps and 2 cm for courtship long chirps and intruder small chirps. Distances between the intruding male and the courting male during assessment behavior (Fig. 8C, top) were measured every 40 ms beginning with the appearance of the intruding fish until the eventual approach or attack. These distances, collected from a total assessment time of 923 s, were summarized in a kernel density histogram with Gaussian kernels with a SD of 2 cm.
Based on the results and procedures from Figure 8B, we defined “courting dyads” as pairs in which a male fish chirped at a female within a range of 60 cm.
Attack distances between two males (Fig. 8C, bottom) were determined at the moment a resident male initiated its movement toward an intruding male. This moment was clearly identifiable as the onset of a linear movement of the resident male toward the intruder from plots showing the position of the fish as a function of time.
The distribution of beat frequencies generated by fish present in the electrode array at the same time (Fig. 8E) was calculated from all recordings. The average frequency difference of each pair of fish simultaneously detected in the recordings was compiled into a kernel density histogram with a Gaussian kernel with a SD of 10 Hz. Similarly, for courtship and aggressive behavior (Fig. 8F,G), the mean frequency differences were extracted for the duration of these interactions.
Electric fields.
For an estimation of EOD amplitude as a function of distance, histograms of envelope amplitudes from all electrodes of the array were computed as a function of distance between the electrodes and the estimated fish position. For each distance bin in the range of 20–100 cm, the upper 95% percentile of the histogram was determined and a power law was fitted to these data points. Gymnotiform electroreceptors measure the electric field; that is, the first spatial derivative of the EOD amplitudes, as shown in Figure 8A.
Breeding monitoring setup
In the laboratory breeding study, we used the brown ghost knifefish A. leptorhynchus, a close relative of A. rostratus (de Santana and Vari, 2013). The two species share many similarities: most chirps produced by both species are “small chirps” that in A. leptorhynchus have been classified as type 2 chirps (Engler and Zupanc, 2001), females of both species additionally generate small proportions of “long chirps” similar to the type 4 chirps classified for A. leptorhynchus males, and both species show the same sexual dimorphism in EODf.
The laboratory setup for breeding A. leptorhynchus consisted of a tank (100 × 45 × 60 cm) placed in a darkened room and equipped with bubble filters and PVC tubes provided for shelter. Water temperature was kept between 21°C and 30°C. The light/dark cycle was set to 12/12 h. Several pieces of rock were placed in the center of the tank as spawning substrate. EOD signals were recorded differentially using four pairs of graphite electrodes. Two electrode pairs were placed on each side of the spawning substrate. The signals were amplified and analog filtered using a custom-built amplifier (100× gain, 100 Hz high-pass, 10 kHz low-pass; NPI Electronics), digitized at 20 kHz with 16 bit (PCI-6229, National Instruments), and saved to hard disk for offline analysis. The tank was illuminated at night with a dozen infrared LED spotlights (850 nm, 6 W, ABUS TV6700) and monitored continuously (Movie 4) with two infrared-sensitive high-resolution video cameras (Logitech HD webcam C310, IR filter removed manually). The cameras were controlled with custom-written software (https://github.com/bendalab/videoRecorder) and a timestamp for each frame was saved for later synchronization of the cameras and EOD recordings. Six fish of A. leptorhynchus (three male, three female; imported from the Río Meta region, Colombia) were kept in a tank for 1 year before being transferred to the recording tank. First, fish were monitored for ∼1 month without external interference. We then induced breeding conditions (Kirschbaum and Schugardt, 2002) by slowly lowering water conductivity from 830 to ∼100 μS/cm over the course of 3 months by diluting the tank water continuously with deionized water. The tank was monitored regularly for the occurrence of spawned eggs.
Results
We recorded the EODs of weakly electric fish in a stream in the Panamanian rainforest by means of a submerged electrode array at the onset of their reproductive season in May of 2012 (Fig. 1C, Fig. 1-1, Movie 1). Individual gymnotiform knifefish, A. rostratus, were identified and their movements tracked continuously based on the species- and individual-specific frequency of their EOD (EODf ≈580–1050 Hz). In these recordings, we detected several types of “chirps” emitted during courtship and aggression (Fig. 1B). This approach allowed us to reconstruct social interactions in detail (Fig. 2, Movie 2 and 3) and to evaluate the associated sensory scenes experienced by these fish in their natural habitat.
Figure 1-1
Electrocommunication in the wild
We focused on two relevant communication situations, courtship and aggressive dyadic interactions. In total, we detected 54 episodes of short-distance interactions that we interpreted as courtship (see below) between low-frequency females (EODf < 750 Hz, n = 2) and high-frequency males (EODf > 750 Hz, n = 6) (Meyer et al., 1987), occurring in 2 of 5 nights. Courting was characterized by extensive production of chirps (Fig. 2A, 3) by both males and females, with up to 8400 chirps per individual per night (Fig. 4). Most chirps were so-called “small chirps,” characterized by short duration (<20 ms) EODf excursions of <150 Hz and a minimal reduction in EOD amplitude (Engler and Zupanc, 2001) (Figs. 1B, 3). Only females emitted an additional type of chirp in courtship episodes, the “long chirp” (Figs. 1B, 3), with a duration of 162 ± 39 ms (n = 54), a large EODf excursion of ∼400 Hz, and a strong decrease in EOD amplitude (Hagedorn and Heiligenberg, 1985). Per night and female, we observed 9 and 45 long chirps, respectively, generated every 3–9 min (first and third quartile) between 7:00 P.M. 1:00 A.M. (Fig. 4A). Occasionally, courtship was interrupted by intruding males, leading to aggressive interactions between resident and intruder males (see below).
Courtship chirping
Roaming males approached and extensively courted females by emitting large numbers of small chirps (Fig. 4A). Courtship communication was highly structured, with female long chirps playing a central role. Long chirps were preceded by persistent emission of small chirps by the male with rates of up to 3 Hz (Fig. 5A,C, and Fig. 5-2). Immediately before the long chirp, the female small-chirp rate tripled from below 1 to ∼3 Hz within a few seconds. The male chirp rate followed this increase until the concurrent high-frequency chirping of both fish ceased after the female long chirp. These chirp episodes were characterized by close proximity of the two fish (<30 cm; Fig. 5B,D). Long chirps were consistently acknowledged by males with a doublet of small chirps (Fig. 3) emitted 229 ± 31 ms after long chirp onset (n = 53 measured in 5 pairs of interacting fish; Fig. 4A). The two chirps of the doublet were separated by only 46 ± 6 ms, 7-fold shorter than the most prevalent chirp intervals (Fig. 5-1). Finally, the female often responded with a few more loosely timed small chirps ∼670 ± 0.182 ms after the long chirp (time of first chirp observed in n = 33 of the 40 episodes shown in Fig. 5-2). The concurrent increase in chirp rate, its termination by the female long chirp, the male doublet, and the final response by small chirps of the female stood out as a highly stereotyped communication motif that clearly indicates fast interactive communication.
Figure 5-1
Figure 5-2
Males echo female chirps
On a subsecond timescale, male chirping was modulated by the timing of female chirps (Fig. 6A,C). After a female small chirp, male chirp probability first decreased to a minimum at ∼75 ms (significant in 4 of 5 pairs of fish) and subsequently increased to a peak at ∼165 ms (significant in 4 of 5 pairs of fish). In contrast to males, females did not show any echo response (Fig. 6B,D); they timed their chirps independently of the males' chirps.
Competition between males
A second common type of electrocommunication interaction observed in our field data was aggressive encounters between males competing for access to reproductively active females. These aggressive interactions were triggered by intruding males that disrupted courtship of a resident, courting dyad. Intruding males initially often lingered at distances larger than 70 cm from the courting dyad (8 of 16 scenes, median duration 58.5 s; for examples, see Fig. 2A, Movie 2), consistent with assessment behavior (Arnott and Elwood, 2008). Resident males detected and often attacked intruders over distances of up to 177 cm, showing a clear onset of directed movement toward the intruder (Fig. 2C, Movie 2). In 5 of 12 such situations, a few small chirps indistinguishable from those produced during courtship were emitted exclusively by the retreating fish (Fig. 4A). The distances at which resident males started to attack intruders ranged from 20 to 177 cm (81 ± 44 cm, n = 10; Fig. 2B, Movie 3). At the largest observed attack distance of 177 cm, the electric field strength was estimated to be maximally 0.34 μV/cm (assuming the fish were oriented optimally), a value close to minimum behavioral threshold values of ∼0.3–0.1 μV/cm measured in the laboratory at the fish's best frequency (Bullock et al., 1972; Knudsen, 1974). We observed a single rise: a slow, gradual increase in EODf (Zakon et al., 2002) emitted by a retreating intruder fish.
Synchronization of spawning
We investigated the role of the female long chirp in a breeding experiment in the laboratory (Kirschbaum and Schugardt, 2002) by recording and videotaping a group of 3 males and 3 females of the closely related species A. leptorhynchus (de Santana and Vari, 2013) continuously over >5 months. Scanning >1.3 million emitted chirps, we found 76 female long chirps embedded in communication episodes closely similar to those observed in A. rostratus in the wild (cf. Figs. 7B, 3). Eggs were only found after nights with long chirps (six nights). The number of eggs found corresponded approximately to the number of observed long chirps, supporting previous anecdotal findings that Apteronotus females spawn single eggs during courtship episodes (Hagedorn and Heiligenberg, 1985). The associated video sequences triggered on female long chirps show that, before spawning, females swim on their side close to the substrate, such as a rock or a filter, whereas the male hovers in the vicinity of the female and emits chirps continuously (Movie 4). In the last seconds before spawning, the female starts to emit a series of chirps, whereupon the male approaches the female. A fraction of a second before the female emits its long chirp, the male pushes the female and retreats almost immediately afterward (Fig. 7). It seems highly likely that this short episode depicts the synchronized release of egg and sperm.
Statistics of natural stimuli
In a final step, we deduced the statistics of natural electrosensory stimuli resulting from the observed communication behaviors of A. rostratus to be able to relate it to the known physiological properties of electrosensory neurons in the discussion. Superposition of a fish's EOD with that of a nearby fish results in a periodic amplitude modulation, a so-called beat. Both frequency and amplitude of the beat provide a crucial signal background for the neural encoding of communication signals (Benda et al., 2005; Marsat et al., 2012; Walz et al., 2014). The beat frequency is given by the difference between the two EODfs and the beat amplitude equals the EOD amplitude of the nearby fish at the position of the receiving fish (Fotowat et al., 2013).
The EOD amplitude and thus the beat amplitude decay with distance. We measured this decay directly from the data recorded with the electrode array (Fig. 8A). The median EOD field amplitude at 3 cm distance was 2.4 mV/cm (range 1.4–5.1). The electric field decayed with distance according to a power law with exponent 1.28 ± 0.12 (n = 9). This is less than the exponent of 2 expected for a dipole, because the water surface and the bottom of the stream distort the field (Fotowat et al., 2013). Small and long chirps emitted during courtship and small chirps emitted by retreating intruder males occurred at small distances of <32 cm (Fig. 8B). In contrast, two behaviors involving intruding males occurred at large distances (Fig. 8C). First, intruding males initially often lingered at distances larger than 70 cm from the courting dyad (n = 8, median duration 58.5 s; for examples, see Fig. 2A, Movie 2), consistent with assessment behavior (Arnott and Elwood, 2008). Second, the distances at which resident males started to attack intruders ranged from 20 to 177 cm (81 ± 44 cm, n = 10; Fig. 2B, Movie 3). At the largest observed attack distance of 177 cm, we estimated the electric field strength to be maximally 0.34 μV/cm, assuming the fish were oriented optimally.
All courtship chirping occurred at high beat frequencies (205–415 Hz for the five pairs in which the female emitted long chirps; Figs. 8F, 4B). High beat frequencies were not a rare occurrence, as the probability distribution of 406 beat frequencies measured from encounters in five nights show (Fig. 8E). From these, the 183 male–female encounters resulted in beat frequencies ranging from 99 to 415 Hz. Conversely, same-sex interactions resulted in low beat frequencies up to 245 Hz (Fig. 8E). Encounters between females were more frequent than between males (187 female vs 36 male encounters). Female EODfs ranged from 585 to 748 Hz and resulted in observed beat frequencies from 1 to 142 Hz. Beat frequencies of 49 Hz were the most frequent among the females (n = 187). Conversely, male EODfs spanned a much larger range from 776 to 1040 Hz, resulting in a broad and flat distribution of beat frequencies spanning 12 to 245 Hz (peak at 98 Hz, n = 36). This includes the range of beat frequencies observed at aggressive male–male interactions (Fig. 8G).
Discussion
We recorded movement and electrocommunication signals in a wild population of the weakly electric fish, A. rostratus, in their natural Neotropical habitat. A stereotyped pattern of interactive chirping climaxed in a special long chirp emitted by the female that we identified as a synchronizing signal for spawning. Courtship chirping was characterized by concurrent increases in chirp rate of both males and females on a tens-of-seconds time scale and by echo responses by the males on a 100 ms time scale. Courtship chirping occurred at distances below 32 cm and on high beat frequencies of up to 415 Hz. In contrast, aggressive interactions between males occurred at beat frequencies below ∼200 Hz and often at distances larger than half a meter.
Communication in the wild and in the laboratory
Our observations of male echo responses to female chirps (Fig. 6A,C), precisely timed chirp doublets in response to female long chirps (Fig. 3), immediate behavioral reactions of males to female long chirps (Fig. 7, Movie 4), and females slowly raising their chirp rate in response to male chirping and responding to the male's chirp doublet (Fig. 5 and Fig. 5-2) clearly qualify chirps as communication signals in natural conditions. Laboratory studies have found echo responses on similar (Hupé and Lewis, 2008) or slower (Zupanc et al., 2006; Salgado and Zupanc, 2011; Metzen and Chacron, 2017) time scales exclusively between males. Small chirps have been suggested to deter aggressive behavior (Hupé and Lewis, 2008). This is consistent with our observation of a submissive function of male–male chirping. The number of chirps generated in these aggressive contexts is, however, much lower (1–10 chirps in 5 of 9 pairings; Fig. 4) compared with encounters staged in laboratory tanks (∼125 chirps per 5 min trial; Hupé and Lewis, 2008). Our field data do not support a function of chirps as signals of aggression and dominance (Triefenbach and Zakon, 2008). In particular, the restricted space in laboratory experiments may explain these differences.
In so-called “chirp chamber” experiments, in which a fish is restrained in a tube and is stimulated with artificial signals mimicking conspecifics, small chirps are predominantly generated by males at beat frequencies well below ∼150 Hz, corresponding to same-sex interactions (Bastian et al., 2001; Engler and Zupanc, 2001). In contrast, in our observations of courting fish in the field and in the laboratory, both male and female fish almost exclusively chirped in male–female contexts at beat frequencies above ∼200 Hz (Fig. 4B).
Electric synchronization of spawning by courtship-specific chirps
Our results provide strong evidence that female long chirps are an exclusive communication signal for the synchronization of egg and sperm release for external fertilization as has been suggested by Hagedorn and Heiligenberg (1985). The female long chirp was the central part of a highly stereotyped communication pattern between a courting dyad (Figs. 3, 5, and 5-2). Fertilized eggs were found at the locations of male–female interactions and only when the female had produced long chirps in the preceding night. The period immediately before the female long chirp was characterized by extensive chirp production by the male (Fig. 5). Video sequences triggered on female long chirps clearly demonstrated the special role of the female long chirp (Fig. 7, Movie 4). The videos also show that, in the seconds before emission of the long chirp, the fish were in very close proximity. Therefore, additional cues such as high beat amplitudes and touch might also play a role in synchronization of fertilization.
Robust responses to communication signals
Male echo responses to female chirps occurring reliably within a few tens of milliseconds (Fig. 6A,C), precisely timed chirp doublets (Fig. 3), and long-range assessment and attacks (Fig. 8C) demonstrate that the respective electrocommunication signals are successfully and robustly evaluated by the electrosensory system, as would be expected for communication signals (Wilson, 1975; Endler, 1993). The electrosensory signals arising in these interactions are dominated by beats; that is, amplitude modulations arising from the interference of the individual electric fields.
Two types of tuberous electroreceptor afferents could contribute to the observed behavioral responses in A. rostratus. T-units play an important role in the jamming avoidance response (Bullock et al., 1972; Rose and Heiligenberg, 1985). However, whether and how T-units are able to encode beats with frequencies >20 Hz is not yet known. P-units, the dominant type of tuberous receptors (Carr et al., 1982), encode amplitude modulations of the fish's EOD by modulating their firing rate (Scheich et al., 1973; Bastian, 1981a; Nelson et al., 1997; Benda et al., 2005; Walz et al., 2014). Tuning of P-unit firing-rate modulations, spike time correlations, and stimulus–response coherences to beat frequencies have been characterized up to beat frequencies of 300 Hz by single-unit, dual-unit, and nerve recordings (Bastian, 1981a; Nelson et al., 1997; Benda et al., 2006; Walz et al., 2014). These measures are on average strongest at beat frequencies of ∼30 to 130 Hz (Bastian, 1981a; Benda et al., 2006; Walz et al., 2014; Grewe et al., 2017), covering well the beat frequencies arising from same-sex interactions (Fig. 8G). For higher beat frequencies, firing-rate modulations and related measures decay down to lower values (Fig. 8H). Encoding of low beat frequencies occurring during male–male interactions is thus well understood.
Neglected stimulus frequencies
Only very few studies have looked at P-unit responses to beat frequencies beyond 300 Hz and none has addressed the encoding of chirps beyond 250 Hz. Narrow-band amplitude modulations of up to 400 Hz were shown to evoke sizable stimulus–response coherences (Savard et al., 2011). Based on our findings from this field study, we started to investigate the encoding of high beat frequencies and found significant spike time locking of P-units to beat frequencies up to 500 Hz (Sinz et al., 2017). These data seem to parallel spike time locking to amplitude modulations described in the peripheral vertebrate auditory systems (Joris et al., 2004) and might explain how high beat frequencies are reliably represented, as the courtship behaviors that we observed suggest. Future studies need to explore these coding schemes, in particular with respect to the encoding of chirps occurring on beat frequencies beyond 250 Hz.
The difference between the high beat frequencies that we observed during courtship interactions (205–415 Hz; Figs. 8F, 4B) and the peak of the frequency tuning of the firing rate (Fig. 8H) is unexpected given the many examples of frequency-matched courtship signals in other sensory systems (Rieke et al., 1995; Machens et al., 2005; Kostarakos et al., 2009; Schrode and Bee, 2015). The high beat frequencies result from males having higher frequencies than females (Meyer et al., 1987). In the genus Apteronotus, the presence, magnitude, and direction of EODf dimorphism varies considerably across species and thus is evolutionarily labile (Smith, 2013).
Neglected stimulus amplitudes
The field strength of the EOD, and with it beat amplitude, decays with distance (Fig. 8A). Most of the studies on P-unit coding, including Savard et al. (2011) and Sinz et al. (2017), used rather strong beat amplitudes of >10% of the EOD amplitude. We observed chirp interactions at distances up to 32 cm, corresponding to beat amplitudes of ∼1% (Fig. 8A). Opponent assessment and decision to attack usually occur at even larger distances (Fig. 8C), where the relevant signal amplitudes are much smaller than 1% of the fish's own EOD amplitude. In general, smaller beat amplitudes result in down-scaled frequency tuning curves (Bastian, 1981a; Benda et al., 2006; Savard et al., 2011; Grewe et al., 2017) and reduced phase locking (Sinz et al., 2017). However, encoding of beats and chirps has so far only been studied for amplitudes larger than 1% (Bastian, 1981a; Nelson et al., 1997).
Decoding
P-units converge onto pyramidal cells in the electrosensory lateral line lobe (ELL) (Heiligenberg and Dye, 1982; Maler, 2009). The rate tuning curves of pyramidal cells peak at frequencies similar to or lower than those of P-units (Bastian, 1981b) and their stimulus–response coherences peak well below 100 Hz, but have only been measured up to 120 Hz (Chacron et al., 2003; Chacron, 2006; Krahe et al., 2008). In contrast to the auditory system, where phase locking to amplitude modulations in neurons of the cochlear nucleus is improved relative to auditory nerve fibers (Joris et al., 2004), phase locking in pyramidal cells compared with P-unit afferents is reduced (Sinz et al., 2017). Coding of small chirps by pyramidal cells in the ELL and at the next stage of processing, the torus semicircularis, has so far only been studied at beat frequencies below 60 Hz (Marsat et al., 2009; Marsat and Maler, 2010; Vonderschen and Chacron, 2011; Marsat et al., 2012; Metzen et al., 2016). Therefore, most electrophysiological recordings from the electrosensory system have been biased to low beat frequencies and strong stimulus amplitudes evoking obvious neuronal responses, but overlooking the stimuli relevant for reproduction.
Conclusion
Our observations regarding sex specificity, numbers, and functions of chirps differ substantially from laboratory studies. The fish robustly responded to courtship signals that occurred on beat frequencies that were unexpectedly high given previous, mainly laboratory-based findings on chirping (Smith, 2013; Walz et al., 2013). In addition, male fish initiated attacks at distances resulting in unexpectedly low beat amplitudes. These ranges of stimulus frequencies and amplitudes have been largely ignored by electrophysiological characterizations of the electrosensory system. Our field data thus identify important, but so far neglected, stimulus regimes of the electrosensory system and provide further evidence for the existence of sensitive neural mechanisms for the detection of such difficult sensory signals (Gao and Ganguli, 2015). Our work also points to the limitations of laboratory studies and emphasizes the importance of research in the natural habitat, which opens new windows for understanding the real challenges faced and solved by sensory systems.
Footnotes
This work was supported by the German Federal Ministry of Education and Research (BMBF Bernstein Award for Computational Neuroscience 01GQ0802 to J.B.), the Natural Sciences and Engineering Research Council of Canada (Discovery Grant to R.K.), and the Smithsonian Tropical Research Institute (Short Time Fellowship to J.H.). We thank Hans Reiner Polder and Jürgen Planck from NPI Electronic GmbH for designing the amplifier; Sophie Picq, Diana Sharpe, Luis de León Reyna, Rigoberto González, Eldredge Bermingham, the staff from the Smithsonian Tropical Research Institute, and the Emberá community of Peña Bijagual for logistical support; Fabian Sinz for advice on the analysis; and Leonard Maler, Ulrich Schnitzler, and Janez Presern for comments on the manuscript.
The authors declare no competing financial interests.
- Correspondence should be addressed to either Jan Benda or Jörg Henninger, Institut für Neurobiologie, Eberhard Karls Universität, Auf der Morgenstelle 28E, 72076 Tübingen, Germany, jan.benda{at}uni-tuebingen.de or joerg.henninger{at}posteo.de