Neuroanatomical and neurophysiological mechanisms of acoustic and weakly electric signaling in synodontid catfish

Abstract To what extent do modifications in the nervous system and peripheral effectors contribute to novel behaviors? Using a combination of morphometric analysis, neuroanatomical tract‐tracing, and intracellular neuronal recording, we address this question in a sound‐producing and a weakly electric species of synodontid catfish, Synodontis grandiops, and Synodontis nigriventris, respectively. The same peripheral mechanism, a bilateral pair of protractor muscles associated with vertebral processes (elastic spring mechanism), is involved in both signaling systems. Although there were dramatic species differences in several morphometric measures, electromyograms provided strong evidence that simultaneous activation of paired protractor muscles accounts for an individual sound and electric discharge pulse. While the general architecture of the neural network and the intrinsic properties of the motoneuron population driving each target was largely similar, differences could contribute to species‐specific patterns in electromyograms and the associated pulse repetition rate of sounds and electric discharges. Together, the results suggest that adaptive changes in both peripheral and central characters underlie the transition from an ancestral sound to a derived electric discharge producing system, and thus the evolution of a novel communication channel among synodontid catfish. Similarities with characters in other sonic and weakly electric teleost fish provide a striking example of convergent evolution in functional adaptations underlying the evolution of the two signaling systems among distantly related taxa.

Similarly, alterations of synaptic strength and intrinsic membrane properties in a three-neuron stomatogastric network can result in the generation of gastric mill patterns with similar or different periods and burst durations (Prinz, Bucher, & Marder, 2004). Temporal differences in the courtship calls of pipid frogs (Xenopus) can be explained, in part, by the intrinsic properties of premotor neurons (Barkan, Kelley, & Zornik, 2018), while divergent spectral features appear to be intrinsic to the larynx (Kwong-Brown et al., 2019). While these studies demonstrate that species diversity in an acoustic signal can depend on both central and peripheral properties within a single genus, it remains an open question as to what might accompany the evolution of novel behaviors. One strategy to best identify such characters would be to compare closely related species having the same muscle accomplishing two different functions. The protractor motor system of synodontid catfish offers such an opportunity because it is used for the production of sound, weak electric fields, or both, depending on the species (Baron, Morshnev, Olshansky, & Orlov, 1994;Boyle, Colleye, & Parmentier, 2014;Hagedorn, Womble, & Finger, 1990;Orlov & Baron, 2005;Orlov, Baron, & Golubtsov, 2017).
The protractor muscle of synodontids inserts on the Müllerian ramus, a vertebral process that ends in a plate-like structure lying on the swim bladder wall; together referred to as an elastic spring apparatus (ESA) (Parmentier & Diogo, 2006). Comparative studies strongly suggest that the ancestral function of the ESA is sound production . Although electrogenic synodontids retain an ESA, their protractor muscle fibers have many fewer myofibrils , like the myogenic electric organ of other genera of weakly electric fish (e.g., Bass, Denizot, & Marchaterre, 1986;Bennett, 1971). The protractor muscles are innervated ipsilaterally by a protractor motor nucleus containing dense clusters of motoneurons that extends along the midline from the caudal hindbrain into the rostral spinal cord (Hagedorn et al., 1990;Ladich & Bass, 1996).
Populations of protractor premotor neurons are located in the dorsolateral part of the motor nucleus (Type I) and close to the rostral end of the motor nucleus, lateral to the medial longitudinal fasciculus (Type 2) (Ladich & Bass, 1996). Ladich and Bass (1996) also report extensive connections between this premotor-motor circuit and a region comparable in location to the vocal prepacemaker nucleus in the toadfish vocal system (Bass, Marchaterre, & Baker, 1994;Chagnaud, Baker, & Bass, 2011;Chagnaud & Bass, 2014), suggesting a hindbrain vocal network comparable in organization to that of toadfishes.
F I G U R E 1 Swim bladder sounds and electric discharges (EDs) recorded from Synodontis grandiops and S. nigriventris, respectively. (a) Spectrogram (top) and waveform (middle and bottom at two timescales) of a pulse train produced by a S. grandiops (S.g.). (b) Spectrogram (top) and waveform (middle and bottom at two timescales) of a tonal sound produced by a S. grandiops (S.g.). (c) Spectrogram (top) and waveform (middle and bottom at two timescales) of an ED burst produced by a S. nigriventris (S.n.). Water temperature: 26 ± 1 C [Color figure can be viewed at wileyonlinelibrary.com] To uncover adaptive features associated with the production of either sound or electric discharge (ED), we investigated neuroanatomical and neurophysiological characters of the ESA communication system in two species that show only one type of signaling mechanism; the sound-producer Synodontis grandiops and the ED-producer Synodontis nigriventris (Figure 1). We investigated the signal output (sound or ED) and morphology of the ESA, and recorded electromyograms (EMGs) of the protractor muscles that lead to such different signals. To understand how protractor muscle activity might be controlled, the hindbrain network, as well as the intrinsic membrane properties of protractor motoneurons, were studied. We found that the general organization of the ESA and its associated premotor-motor network was similar in both sonic and ED species; however, multiple ESA and premotor-motor morphological characters differed. Neurophysiological measures further indicated that the motoneuron populations in both species control fast-synchronous activation of the paired protractor muscles; however, there were species-specific patterns that could account for differences in sound and electric discharge pulse repetition rate.

| Animals
Twenty-five S. grandiops (standard length, SL: 54 to 108 mm) purchased from "Les aquariums de Marbais" (Belgium) and 33 S. nigriventris (SL: 42 to 71 mm) purchased from "EFS Nürnberg" (Germany) were maintained at either Liège University or the Ludwig-Maximilians-University (LMU) in monospecific tanks of~200 L (photoperiod: 12:12 hr L:D; water temperature: 26 ± 1 C). The tanks were supplied with numerous hiding places and the fish were fed daily with commercial fish food. While not all animals were sexed, observations of males and females did not reveal any obvious sexual dimorphism in the ESA. As the animals were acquired from the aquarium trade, their age was undetermined. All experimental procedures were approved by the Institutional Animal Care and Use Committee of the University of Liège (protocols 1,970 and 2,110) and the Regierung von Oberbayern (55.2-1-54-2532-13-2016).

| Sound analysis
Fish from both species were recorded following the protocol described in Boyle et al. (2014). During each trial, the simultaneous recording of sounds and EDs were performed. Synodontids also generate high-frequency stridulation sounds by rubbing their pectoral spine against their pectoral girdle (Parmentier et al., 2010). These sounds were produced by both species, but were not analyzed as the focus here is on the shared motor system activating swim bladderassociated muscles that are either sonic or electrogenic. Sounds were recorded with an HTI-Min 96 hydrophone (−186.4 dBV re 1 μPa, frequency response 2 Hz-30 kHz; High Tech Inc., Long Beach, MS) or an AS-1 hydrophone (−208dBV re 1 μPa, frequency response 1 Hz to 100 kHz; Aquarian Audio, Anacortes, WA) with a PA-4 hydrophone preamplifier (Aquarian Audio), while EDs were recorded with two Teflon-coated silver electrodes (4 cm exposed tips) separated by 25 cm. These electrodes were connected to a differential amplifier (A-M Systems Model 1,700) that filtered (bandpass: 10 Hz-10 kHz; notch filter: on) and amplified the signal (10 k fold). The hydrophone and the amplifier were connected to an external sound card (Creative model SB0270; Creative Labs, Singapore or UltraLite-mk3; MOTU; Cambridge, MA), and the signals were recorded on a laptop using Adobe Audition 2.0 (Adobe, San Jose, CA) or audacity 2.0.5 (http:// sourceforge.net/projects/audacity/). Swim bladder sounds were recorded from seven S. grandiops in two different glass tanks (tank 1: 60 × 29 cm, water depth kept at 20 cm; tank 2: 108 × 48 cm, water depth kept at~30 cm) and the EDs from four S. nigriventris in a plastic tank (45 × 27 cm, water depth kept at~20 cm). For both types of signals, we measured the signal duration, the number of oscillations, the oscillation period, and the peak frequency. As EDs had a high signal-to-noise ratio (SNR), a custom written semi-automated analysis (written in Igor; Wavemetrics, Lake Oswego, OR) was used to determine the above-described signal parameters. The sounds (50 tonal and 142 pulses) were analyzed manually using Adobe audition because they had a lower SNR and the background noise had a variety of peaks. 170 HD camera, Leica Microsystems GmbH, Wetzlar, Germany). The midline length, that is, the distance between the two main insertion points of the protractor muscle following the midline of the muscle in a lateral view, and maximal thickness of the tissue in a dorsal view were measured in ImageJ (Wayne Rasband, National Institutes of Health). The Müllerian ramus was imaged and its length, plate surface area, and stem and stem process lengths were measured following the same procedure.

| Neuronal network identification
To visualize the neuronal network, fish were first anesthetized with 0.025% benzocaine (Sigma Aldrich Chemie GmbH, Munich, Germany) or 0.02% tricaine methanesulfonate (MS 222, Sigma Aldrich BVBA, Overijse, Belgium) dissolved in aquarium water. Long-term anesthetic (bupivacaine, 0.25%) was applied with a soaked tissue placed on top of the surgical site and the protractor muscle was exposed. The nerve innervating the protractor muscle was cut at the level of the muscle and its proximal end-labeled by direct application of crystals of either dextran-rhodamine in five S. grandiops (SL: 67 to 88 mm) and four S. nigriventris (46 to 57 mm) or neurobiotin in two S. grandiops (54 to 58 mm) and three S. nigriventris (46 to 58 mm). After a survival time of two to 4 days, the fish were euthanized with an overdose of either
Amplifying the signal 100-fold generally provided EMGs with good SNRs and prevented clipping. Total duration, number of pulses, and pulse period were analyzed for three to five EMGs per fish and compared to the features of the associated sound or ED. For each S. grandiops, three EMGs associated with single-pulse sounds were also analyzed. EMG amplitudes were not investigated because the distance between the two tips of an electrode, the extent of the electrode tips exposed, and the electrode position in the muscle were not rigorously controlled.

| Intrinsic properties of protractor motoneurons
To investigate the intrinsic membrane properties of the motoneurons innervating the protractor muscle, fish were deeply anesthetized with benzocaine and then weighed and measured before being placed in an ice-filled dish. The braincase was opened, and the nuchal shield and underlying muscles were removed. The brain and the rostral part of the spinal cord were removed and placed in an ice-cold solution (in mM: 120 sucrose, 25 NaCl, 27NaHCO3, 2.5 KCl, 1.25 NaH2PO4, 3 MgCl2, 0.1 CaCl2, 25 glucose, 0.4 ascorbic acid, 3 myoinositol, and 2 Na-pyruvate). Care was taken to remove the dura mater. The brain was then embedded in low melting agar (2% in 0.1 M PB) and cut on a vibratome (Vibrating Microtome 7,000 smz-2; Campden Instruments Ltd., Loughborough, Leics, England) at a thickness of 200 μm in the sagittal plane while submerged in the ice-cold solution. Slices were incubated at room temperature in a solution like that used above for dissection but with 125 mM NaCl, 1.2 mM CaCl2, and 1 mM MgCl2 and no sucrose, and oxygenated with 95% O 2 and 5% CO 2 .
Glass recording electrodes were filled with a solution composed of (in mM) 145 K-gluconate, 4.5 KCl, 15 HEPES, 2 Mg-ATP, 2 K-ATP, 0.3 Na-GTP, and 7.5 Na2-phosphocreatine adjusted by adding KOH to pH 7.25. Data were acquired in current-clamp mode at a sampling frequency of 100 kHz and low pass filtered at 3 kHz.
Forty-seven of these motoneurons in three S. grandiops (SL: 83 and 108 mm) and all the S. nigriventris were successfully filled with Alexa (488 or 546) hydrazide, imaged at high resolution under a confocal laser microscope (Leica microsystems, Wetzlar, Germany), and the image stacks used to reconstruct motoneuron somata and neurites in three dimensions using Neurolucida 360 software (MBF Bioscience, Williston, VT). The resting membrane potential was determined as well as passive membrane properties (input resistance, capacitance, time constant Tau), which were obtained from a − 50 pA current pulse of 0.6 s duration (1.1 s inter-stimulation interval). Action potential (AP) firing was investigated by applying a depolarizing current pulse of varying amplitude (duration 0.6 s) until spike adaptation occurred. The rheobase (i.e., a minimal amount of current to evoke a single AP) was determined with the same stimulus (the holding current was not subtracted from the rheobase as it was similar in both species), but at current steps of 5 pA at a 1.1 s inter-stimulus interval. Once the rheobase was found, we repeatedly stimulated the neurons 20 times at the rheobase current threshold to determine AP firing reliability and first AP latency. Action potential features such as amplitude, halfwidth, afterhyperpolarization (AHP) amplitude, AHP half-width, and afterdepolarization amplitude were obtained from individual APs that were elicited by a 0.38 ms duration pulse with a constant current intensity set by the experimenter and a 50 ms inter-stimulation interval, as previously done in the auditory system (e.g., Ammer, Grothe, & Felmy, 2012). Stimulus duration was increased by 0.1 ms steps until an AP was fired.

| Statistical analysis
For standardization of size, ESA morphometric data were divided by the length of the Müllerian ramus, protractor nucleus length was divided by the respective fish's SL, and neuron counts and measurements were divided by the length of the corresponding nucleus. Outliers in the data collected during the different patch-clamp protocols were identified with the online GraphPad calculator (https://www. graphpad.com/quickcalcs/Grubbs1.cfm) and removed from the graphs and statistics. Principal component analyses obtained in the software Past 3.15 (Hammer 1999-2017) were used to explore species distribution in two-dimensional spaces using multivariate data samples. Significance tests were combined with estimation statistics to improve our interpretation of the results. For all significance tests, the null hypothesis was that S. grandiops and S. nigriventris did not differ in a measurement. We tested this hypothesis with two-sample Hotelling's T 2 tests for multivariate comparisons using the "Hotelling" package of R. Subsequent comparisons of univariate data were performed with Student t tests (or U Mann-Whitney when the assumption of normality was not met) using Prism 5 and alpha levels rectified with the Sequential Bonferroni Correction. The effect size (mean differences and standardized Hedges' g) and the lower and upper bound for a 95% confidence interval were calculated online (http://www.

| Swim bladder sounds and electric discharges
During our behavioral monitoring sessions, both species produced stridulation sounds with their pectorals (see: Parmentier et al., 2010).
However, S. grandiops also produced swim bladder-associated sounds, while S. nigriventris produced only EDs (Figure 1). Synodontis grandiops produced at least two types of swim bladder-associated sounds (Table 1) Hz. An individual emitted both types of sounds as a single event or as a part of a train of successive events with an inter-event interval ≤ 1 s. The interevent intervals were highly variable, ranging from a dozen to hundreds of milliseconds. However, for some trains, single events comprised of up to 10 pulses lasted between 31 and 197 ms and had a pulse period that was relatively short and stable (22.9 ± 6.6 ms, N = 41). We categorized these individual events as grunt sounds.
The wave-like EDs of S. nigriventris were more stereotyped and only emitted as bursts with a stable oscillation period of 9.9 ± 1.6 ms ( Figure 1c; Table 1 Comparison between the tonal sounds and electric organ discharge (ED) bursts recorded from Synodontis grandiops and S. nigriventris   Figure 3b) and a longer stem, but shorter stem process (Figure 3c), while its PM was thinner (Figure 3d).
The absolute values of Hedges' g for these variables varied between 3.83 and 6.39, which suggested extremely large interspecific differences. Using the effect size obtained from the mean differences, we estimated that the plate surface and stem processes were 91 and 64% smaller, respectively, in S. nigriventris, while its stem was 76% longer and its PM was 61% thinner (Table S2). Because of the Bonferroni correction, muscle mass was not tagged as significantly lighter in S. nigriventris. However, the effect sizes suggested a very large species effect (Hedges' g: −2.76); using the effect size obtained from the mean differences the PM was 75% lighter in S. nigriventris (Table S2).
The plate thickness and the muscle midline length, on the other hand, were similar in both species (see Table S2 for the univariate tests, effect sizes and the associated bounds for 95% confidence interval).

| EMGs of protractor muscle
As shown for a tonal sound and ED (top trace; Figure 4a Table 2). A short delay was noticed between the activation potentials in the left and right PM and the associated sound or ED pulse. This delay was longer in S. grandiops than S. nigriventris (for the left and right protractor muscle, respectively; S. g.: 6.7 ± 1.2 ms and 6.8 ± 1.1 ms; S. n.: 1.5 ± 0.8 ms and 1.3 ± 0.9 ms). Similarly, single pulse sounds of S. grandiops were preceded by a single activation potential in both PMs (not shown). Here, the delays between the activation potential in left and right PM and the sound were 6.5 ± 1.9 ms and 7.3 ± 0.5 ms, respectively. The similar temporal patterns for the EMG and signal imply that both PMs are activated simultaneously and that every activation potential is responsible for a single muscle contraction in S. grandiops and ED in S. nigriventris. S. nigriventris (S.n.) (Figure 5d). Normalization of the data using body size showed that S. grandiops had significantly more protractor motoneurons than S. nigriventris (Figure 5e left, Table S3). The effect size for this variable was large (Hedges' g: −1.72) and suggested that S. nigriventris had 65% fewer motoneurons (Table S3)

| Premotor neurons
Three populations of premotor neurons were identified in both species (PN1-3, Figure 6a,b) following the labeling of one protractor muscle with neurobiotin, a tracer that is sufficiently small to pass through gap junctions (Bass et al., 1994). As previously described (Ladich & Bass, 1996), Type 1 neurons (PN1) were located in the PN (Figure 6c). PN2 neurons were located ventrolateral to the rostral PN (also see Ladich & Bass, 1996) in both species and appeared to be far more numerous in S. grandiops (Figure 6e). Fibers projecting rostrally from the PN2 could be followed to a few lightly labeled PN3 neurons forming a small bilateral nucleus in the ventrolateral reticular formation (Figure 6a,b,f). Only a few neurons were labeled in PN3, but a dense mesh of fibers and putative terminal boutons were observed. Contralaterally labeled fibers must have originated from PN1 and/or PN2 ( Figure 6c).
While the general pattern of the premotor organization was shared between the two species, neurobiotin tracings showed some differences. In S. nigriventris, many protractor motoneuron and PN1 somata (88.7 ± 28.5 cells) were also labeled in the contralateral PN ( Figure 6d), though more lightly labeled than on the ipsilateral side.
Contralateral labeling appeared extremely rare (3.9 ± 1.7 cells) in S. grandiops. This most likely explained why the size distribution of the somata showed three peaks for the contralateral labeling in S. grandiops, while only two peaks were observed for the other neurobiotin labeling (Figure 6d). Finally, we found transneuronal labeling of some PN3 neurons in S. nigriventris, but not in S. grandiops. The size of PN3 neurons (12.1 ± 2.6 μm; N = 26) was similar to that of PN1 and PN2. Despite these small differences, our experiments showed that the protractor muscle of these two species is controlled by the same motor-premotor circuit.

| Intrinsic properties of protractor motoneurons
Having established the basic anatomical pattern, we next asked whether neurophysiological differences could be observed at the level of individual neurons. Patch-clamp recordings from 14 motoneurons in three S. grandiops (S. g.) and 12 in four S. nigriventris (S. n.) were highly stable with low holding current (mean ± SD: −5 ± 56 pA).

Motoneurons in both species showed no spontaneous activity and
APs could only be evoked with considerable amounts of current (hundreds of pA) shortly after current injection onset. 1.9 ± 0.9 ms, respectively) were significantly larger in S. nigriventris (Figure 8c,d). Effect sizes for the input resistance and the time constant were large (Hedges' g: 1.38 and 1.59, respectively) and the "Mean Difference" method estimated that they increased by 114 and 120%, respectively. Those differences were, however, not reflected in motoneuron capacitance (Figure 8e, Table S4).  Table S4). While the rheobase was relatively high in both species, it was significantly lower in S. nigriventris compared to S. grandiops (460 ± 328 pA and 1,705 ± 468 pA, respectively) ( Figure 9d, Table S4). The standardized effect size was very large (Hedges' g: −2.94) and the rheobase was estimated to be 68% lower in S. nigriventris (Table S4). Motoneuron firing precision was tested by measuring the AP latency for repeated stimulus applications at the rheobase threshold current (Figure 9b). Both species showed relatively short (28.9 ± 23.4 ms in S. grandiops and 32.5 ± 12.4 ms in S. nigriventris) latencies (Figure 9d) that did not differ statistically.
Again, the effect size was negligible (Table S4).
Action potential latency to a current injection mimicking a strong synaptic input was also measured (Figure 10a,b) and, although it was very short in both species (1.9 ± 0.24 ms in S. grandiops and 2.57 ± 0.6 ms S. nigriventris), it differed significantly at their respective thresholds ( Figure 10c, Table S4). Interestingly, the AP was fired in the downward slope of the membrane potential following current injection in S. grandiops (see Figure 10a), which might be due to either the membrane properties or a longer distance of the axon hillock from the soma in S. grandiops compared with S. nigriventris. S. grandiops and S. nigriventris had similar AP amplitudes (84.4 ± 9.6 mV and 84 ± 6 mV, respectively) and half-widths (0.21 ± 0.03 ms and 0.19 ± 0.03 ms, respectively) ( Figure 10c, Table S4). The AHP was significantly larger (−24.3 ± 4.3 mV and − 14.8 ± 8.7 mV, respectively) and longer (0.72 ± 0.28 ms and 0.29 ± 0.07 ms, respectively) in S. nigriventris compared with S. grandiops (Figure 10c, Table S4). The amplitude of the afterdepolarization (ADP), on the other hand, was larger (1 ± 0.3 mV and 3.3 ± 1.9 mV, respectively) in S. grandiops ( Figure 10c, Table S4). Action potential features that showed interspecific differences also had a large (|Hedges' g| > 1.32) standardized effect size (Table S4). Among them, the largest Hedges' g (2.08) was obtained for the half-width of the AHP which was estimated to be 145% larger in S. nigriventris (Table S4).

| DISCUSSION
Comparative studies suggest that sound production is the ancestral character state for the protractor motor system of synodontid catfish (see Boyle et al., 2014). The results presented here would then suggest that both anatomical and neurophysiological modifications in the protractor muscle and its associated hindbrain premotor-motor network underlie the evolution of a novel, weakly electric signaling channel among synodontids. The goal of this study was to identify characters linked to being either sonic or electrogenic, or common to both behaviors. We recognize, however, that limiting our approach to a single sonic-only and a single electrogenic-only synodontid species cannot account for the full range of interspecific variance. Consequently, the presence and functional significance of the characters that we identified here need to be investigated in other synodontid species to more completely assess their contribution to sonic and/or electrogenic signal production.

| Shared characters between vocal and electrogenic species
Shared characters between the neural control of sonic and electric signaling among synodontids likely relate to similar functional demands (Bass, 1989;Bass & Baker, 1997;Bass & Zakon, 2005). First, electric and vocal signals often require synchronous firing of neurons (Bass, 2014;Bass et al., 1994;Bennett, 1971;Carlson, 2006). Synchronization of neuronal activity has been, in part, attributed to gap junction coupling of neurons, a feature found in electric (Bennett, 1971;Bennett, Pappas, Aljure, & Nakajima, 1967;Bennett, Pappas, Giménez, & Nakajima, 1967;Carlson, 2006;Elekes & Szabo, 1985) and sound-producing (Chagnaud, Zee, & Baker, 2012;Pappas & Bennett, 1966) fish. Our and previous (Ladich & Bass, 1996) transneuronal tracing experiments are highly suggestive of gap junctional coupling in the protractor circuit in both S. grandiops and S. nigriventris. Although the presence of gap junctions in these synodontids seems likely given the extensive labeling that we and Ladich and Bass (1996)  and premotor populations, specific uptake mechanisms for biotinderived compounds at chemical synapses should not be discounted (see discussion in Bass et al., 1994). Conversely, an inherent limitation of this tracing method is that additional premotoneurons connected only with chemical synapses to the motoneurons may not be visualized if there is no such mechanism.
Second, the neural networks for vocal and electric signal production often generate a temporally precise activation pattern (Bass, 2014;Bass & Baker, 1997;Bass & Zakon, 2005). In the frog Xenopus laevis (Yamaguchi, Kaczmarek, & Kelley, 2003), for instance, vocal motoneurons fire only under large depolarization currents and with short onset latencies, features that are well suited to follow rhythmic activity. Vocal motoneurons in midshipman fish (Chagnaud et al., 2012) and both of the synodontid species investigated here did not show spontaneous activity, a characteristic onset firing property, and a low excitability, all features in line with neurons adapted to precise firing.
Lastly, in some species, the peripheral organs generating vocal and electric communication signals follow the fast synchronous oscillations of the neural network (Chagnaud et al., 2012;Rome, Syme, Hollingworth, Lindstedt, & Baylor, 1996;Unguez & Zakon, 1998). Calcium uptake by the sarcoplasmic reticulum is a major limiting factor when it comes to contraction speed in vertebrates (Rome & Lindstedt, 1998). Compared with locomotor muscles, sonic muscles have a similar rate of calcium uptake, but a higher content of sarcoplasmic reticulum allowing them to overcome this limitation (Bass & Marchaterre, 1989;Fawcett & Revel, 1961 (Crampton, 2006). This notion is strengthened by strongly electric catfish, in which only a single pair of electromotoneurons innervate the entire electric organ comprised of millions of electrocytes Janetzko, Zimmermann, & Volknandt, 1987). The lower number of protractor motoneurons in the electric species studied here could also be related to the smaller size of its protractor muscle, which could further relate to the number of muscle fibers/electrocytes. Comparisons with other synodontids (or even mochokid) species producing either swim bladder sounds or EDs would help resolve this matter.
Synodontis grandiops, the sonic species studied here, also had more neurons in PN2 than S. nigriventris. PN3, on the other hand, was not clearly labeled in S. grandiops. Since there are currently no electrophysiological recordings from PN1-3, these differences are difficult to interpret in a functional context. It is also difficult to explain the stronger labeling in S. nigriventris of contralateral motoneurons, premotor neurons, and PN3. The most parsimonious explanation at this point would be that it further improves bilateral, synchronous firing of motoneurons and premotor neurons in S. nigriventris, but there is currently no evidence supporting the hypothesis that electric signaling requires a higher level of bilateral synchronization than the production of swim bladder sounds for any fish species.
Based on our EMG experiments, the higher fundamental frequency, or pulse repetition rate, of the S. grandiops tonal sounds compared with the S. nigriventris EDs is associated with a higher rate of simultaneous-occurring activation potentials in the paired protractor muscles (vs. alternate muscle contraction). In S. grandiops, the action potential latency and AHP were shorter, but the AHP was larger suggesting that the S. grandiops protractor sonic motoneurons are tuned to fire successive action potentials faster than the S. nigriventris S. marmorata, S. schall, and S. caudovitattus) can produce ED bursts and (or exclusively, depending on the authors) EDs composed of a single pulse or a pair of pulses (Baron et al., 1994;Boyle et al., 2014;Hagedorn et al., 1990;Orlov et al., 2017;Orlov, Baron, & Golubtsov, 2015). Due to these differences, recordings of individual motor and premotor neurons during ESA activity are needed to fully understand how motoneurons contribute to ED generation (e.g., see Bass & Baker, 1990;Chagnaud & Bass, 2014;Chagnaud et al., 2012).
A major limitation of this approach compared to the patch-clamp experiment presented here is that so far, we have been unable to elicit fictive calls from anesthetized fish.

| Concluding comments
The contribution of neural mechanisms to the evolution of novel communication channels and behaviors remain largely unexplored (but see Bass & Baker, 1997;Hoke, Adkins-Regan, Bass, McCune, & Wolfner, 2019;Katz, 2016). Synodontid catfish are of great interest in this regard because two signaling mechanisms have evolved from the same neuromuscular system within a single genus. While the interspecific differences in motoneuron physiology and, in turn, EMGs reported here likely contribute to species differences in the pulse repetition rate of the resultant behavioral signals, more species need to be investigated to confirm this hypothesis. It may yet be that properties of the protractor muscle-derived electrocytes contribute to the spectral features of individual pulses as they do in the myogenic electric organ of other teleosts (Bass, 1986;Zakon, Zwickl, Lu, & Hillis, 2008). From a broad comparative perspective, the similarities observed in peripheral and central characters with other sonic and weakly electric teleosts suggest similar selective pressures favoring the evolution of shared mechanisms controlling these signaling behaviors among distantly related taxa.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author upon request.