Simplified neural encoding of social communication reflects lifestyle

As signal production changes through speciation, the sensory systems that receive these signals must also adapt to extract the most relevant information effectively. In a species of weakly electric fish, Apteronotus albifrons we examine the unique neurophysiological properties that support the encoding of electrosensory communication signals that the animal would encounter in social exchanges. We compare our findings to known coding properties of the closely related species, Apteronotus leptorhynchus, to establish how these animals differ in their ability to encode their distinctive communication signals. While there are many similarities between these two species, we find notable differences leading to relatively poor coding of the details of the chirp structure. As a result, small differences in chirp properties are poorly resolved by the nervous system. We performed behavioral tests to relate A. albifrons chirp coding strategies to its use of chirps during social encounters. Our results suggest that A. albifrons do not exchange frequent chirps in non-breeding condition. These findings parallel the mediocre chirp coding accuracy in that they both point to the sparse reliance on chirps in social interactions. Therefore, our study suggests that neural coding strategies in the central nervous system vary across species in a way that parallels the behavioral use of the sensory signals.


INTRODUCTION
frequencies are more typical of encounters between fish of opposite sexes or with large 88 differences in body size. A. leptorhynchus produce Type 1 (big) chirps most often in high 89 frequency beat contexts (Hagedorn and Heiligenberg, 1985;Hupé and Lewis, 2008), while low 90 frequency interactions elicit the production of frequent Type 2 (small) chirps (Hupé & Lewis,91 2008). Chirp coding has been previously well characterized in both Apteronotus leptorhynchus 92 and Eigenmannia viscerens. In both species, chirps cause significant changes in the synchrony of 93 electroreceptor firing, either increasing, or decreasing synchronicity depending on chirp type and 94 beat frequency (Stöckl et al., 2014;Walz et al., 2014). The two categories of signals described 95 above also produce different neural responses in the electrosensory lateral line lobe (ELL), the 96 primary sensory area. Small chirps on low frequency beats cause a synchronized and stereotyped 97 bursting response among the pyramidal cells of the ELL. Due to the encoding strategy and the 98 structure of the signal itself, small variation in these chirps cannot be discriminated (Allen and 99 Marsat, 2018; Marsat et al., 2009). Big and small chirps on high frequency beat lead to 100 heterogeneous and graded responses among pyramidal cells and variation in the signals are 101 accurately discriminated (Allen & Marsat, 2018; Marsat & Maler, 2010). Chirp production in A. 102 albifrons is similar, but the most notable difference between these species' chirps is duration; A.  Maler, 1993). Characterization of frequency tuning in the primary sensory area of A. albifrons 106 show differences from A. leptorhynchus that may be an adaptation for the coding of these 107 particularly long chirps (Martinez et al., 2016) 108 In this study, we examine the coding of conspecific social signals and the underlying 109 basic neural properties to understand if, and how, the sensory system is adapted to the specific 110 characteristics of the communication system of A. albifrons. Additionally, we examine the 111 behavior of chirp production to help understand how the behavioral use of the signals, rather than 112 just their structure, could influence the way they are processed. We compare our findings to the 113 well-studied behavior and physiology of the closely related A. leptorhynchus to identify specific 114 neurophysiological adaptations that reflect differences in the structure and behavioral use of 115 chirps in these two species.

117
Diversity in conspecific chirp responses 118 In multiple gymnotid species the lateral segment (LS) of the ELL serves as the primary 119 location for encoding of communication and social signals (Marsat et al., 2009;Metzner, 1999; 120 Metzner and Juranek, 1997), thus we targeted our recordings to pyramidal cells in that segment. 121 To characterize the pattern of responses of A. albifrons pyramidal cells we played a series of 122 chirps mimicking the natural range of reported A. albifrons chirps. Additionally, for comparison 123 we used a small number of chirps with properties more typical of A. leptorhynchus (Dunlap et 124 al., 1998;Zupanc and Maler, 1993). Detailed descriptions of chirp properties used are located in 125 Table 1. In addition to playing chirps with diverse properties, we also varied the frequency of the 126 beat, presenting chirps on both low (10 Hz) and high (100 Hz) frequency signals. neurons, comparable to A. leptorhynchus (Marsat and Maler, 2010) (Fig 2A). However, this level 151 of accuracy only holds for chirps on the 10 Hz beat. ON-cell performance on 100 Hz is 152 extremely poor, even with high numbers of spike trains included in the analysis (Fig 2B). OFF-153 cells are able to detect chirp occurrence on low frequency beats for all chirp types, although less 154 efficiently than ON-cells ( Fig 2C). On 100 Hz OFF-cells can detect the longest, most intense 155 chirps, but perform poorly detecting smaller chirps ( Fig 2D). Therefore, while chirps on low 156 frequencies are reliably detected, detection sensitivity is poor in high frequency contexts.
We assessed the amount of information carried by the response pattern about chirp 159 properties that could support the discrimination of chirp variants. This analysis is similar to that 160 used for chirp detection, but rather than comparing responses of chirps to beats, we compare 161 responses between chirps. Figure 3 shows discrimination ability for chirps grouped by size (See 162   Table 1; Small chirps: 5, 6, 7, 8; Big chirps: 1, 2, 3, 4, 9, 10) for both ON and OFF-cells on high 163 and low beat frequencies.

164
Chirps of all types are more easily discriminable when presented on 10Hz beats rather 165 than on 100Hz beats (Fig 3). On low frequencies, the all chirps are well discriminated by both 166 ON and OFF cells. Performance for chirps on high frequency beats was more varied (Fig 3B); 167 accurate discrimination was achieved only for chirps with big difference in properties ( Fig 3B   168 and Fig S1) and both ON and OFF cell response allows similar discrimination accuracy. These 169 results are unexpected considering that A. leptorhynchus chirps on high frequency beats can be 170 discriminated well using as few as six neurons, and, conversely,exhibit poor discrimination 171 coding on low frequencies (Marsat and Maler, 2010) where A. albifrons performs best.
172 Furthermore, we do not observe an asymmetry in coding accuracy between ON and OFF cells as 173 observed in A. leptorhynchus. 174 When looking at which chirps are well discriminated, chirp duration and the steepness of 175 the frequency rise seem to be the most discriminable features, while total frequency increase and 176 number of peaks are less influential on coding ( Fig S1). However, discriminating the most easily 177 separated chirps is still prone to error, even when recruiting up to 17 neurons. These data indicate 178 that while qualitative observations would suggest that chirp coding is similar between A.

207
The movement of fish as well as the amplitude changes caused by chirps create low 208 frequency changes in AM contrast, or envelopes (Stamper et al., 2013). During chirps, the AM 209 will be high frequency (tens to hundreds of Hz above the beat frequency) whereas the envelope 210 of the chirp is low frequency (e.g. 100 ms chirp will lead to a ~ 10Hz envelope). Therefore, the 211 coding of chirps' low frequency content must be investigated using envelope stimuli. We probed 212 envelope coding using stimuli consisting of RAMs with AM frequencies between 40Hz and 60 213 Hz, containing envelope frequencies of 0-20 Hz (Middleton et al., 2006). Although pyramidal 214 cells do encode the envelope of these stimuli (Fig 4D), the coding accuracy is low compared  increased firing rate only slightly more in response to A. leptorhynchus chirps than to the beat.

242
There was, however, a small population (n=5) of ON-cells that did reliably burst more in 243 response to A. leptorhynchus small chirps than to the beat (Fig 5A, C). The shortest A. albifrons 244 chirps are most similar to the A. leptorhynchus chirps that elicit bursts, but are still long enough 245 to span more than one beat cycle. Even the bursty neurons did not respond to the shortest A. 246 albifrons chirps with bursts (Fig 5B, D). 247 The low numbers of neurons that burst in response to A. leptorhynchus chirps suggests that there 248 are physiologic differences between these two species in regards to burst coding of 249 communication. We examined the differences between bursty and non-bursty neurons in more 250 detail. The subset of bursty neurons also tend to have broader AM tuning (Fig 5C). While most 251 ON-cells exhibited peak coherence at frequencies between 10-20Hz, these cells had peak    interacting fish. Smaller differences in EODf correspond to higher numbers of chirps ( Fig 7A).

303
Unlike A. leptorhynchus, however, chirping does not appear to be sexually dimorphic. Animals 304 used in behavior experiments were not sacrificed to determine sex, but grouped by EODf into 305 high (>1100Hz) and low (<1100Hz) groups which can correspond to females and males 306 respectively in many populations of A. albifrons (Zakon and Dunlap, 1999). This allowed us to 307 compare chirp rates among higher or lower frequency individuals, as well as by large or small 308 differences in EODf. We observed no differences in chirp production between high or low 309 frequency groups (Fig 7B). Further, chirp rate does not vary by pairing type, nor by relative 310 EODf with each pair under our conditions (Fig 7B). These results replicate previous findings that 311 chirp frequency in this species is not sexually dimorphic (Dunlap et al., 1998).

382
Our study did not test for the effects of neuromodulation on tuning and chirp coding.  (Smith, 2013). It is likely that the effects 390 of neuromodulation due to behavioral state could affect the reception and encoding of these 391 chirps as well, altering sensitivity to chirps and possibly even coding accuracy in response to 392 behavioral need. This may particularly influence the coding of chirps on high frequency beats, 393 which we observed was surprisingly poor. This kind of interaction is more likely to occur in 394 breeding contexts, so it is possible that animals in breeding condition could be better able to 395 detect and discriminate these signals.

397
Trade-offs between specialization and generalization 398 Classical neuroethology dictates that the mode of signal production and mode of signal 399 reception must evolve in synchrony so that senders and receivers never lose the ability to 400 exchange information (Bradbury et al., 2011). There are many examples of specialization of 401 particular aspects of sensory systems to accomplish a highly specialized tasks (Endler, 1992). In 402 the case of communication, sensory tuning for sender-receiver matching has been shown 403 repeatedly. However, the converse may also be true. Over-specialization may come at the cost of 404 reduced sensitivity to more general environmental signals. In such a case, it may be more of peripheral sender-receiver mismatching that may be explained by gains in sensitivity to prey 408 or predator signals to (Mason, 1991;Römer, 2016). Maintaining specificity for conspecific 409 communication may be costly both metabolically (Niven and Laughlin, 2008) and in regards to 410 detecting environmental stimuli apart from communication signals.

411
In A. albifrons we show a general match between signal characteristics, low frequency 412 chirp envelopes, and CNS sensitivity to low frequency signals, but a lack of complex coding that 413 would allow for efficient discrimination of chirp identity in all contexts. For this species, 414 investing fewer resources into the coding of social signals may allow the electrosensory system 415 to focus more broadly. This low frequency tuning may be particularly adaptive for prey location 416 and navigation, both tasks that the electrosensory system must perform, and both tasks that 417 require coding of low frequencies. In such a case, our data argue for a match between neural 418 coding and behavior via a relaxation of sender-receiver matching.   response ( (f)) was constructed from (f) as: .   Twenty-eight behavior trials were conducted in a small tank (27x27x14cm) containing 565 water with conductivity, pH, and temperature matched to the home system, and one shelter tube.

566
The tank was enclosed to block ambient light, lit with infrared lights and all trials were recorded 567 via infrared camera (Logitech HD Pro Webcam C920). 14cm carbon rod electrodes placed 568 diagonally from each other in each corner of the tank recorded electrical activity, which was then 569 amplified (A-M Systems, Model 1700) and recorded using a computer sound card.

570
Stranger fish from different home tanks were selected semi-randomly. Defining physical 571 features (size, markings) and EODf were noted to avoid repeatedly testing the same pair. One 572 fish was selected and allowed to acclimate to the test tank for 20 minutes before the introduction 573 of the second fish. Recording began immediately upon introduction of the intruder fish.

574
Interactions were recorded for five minutes.

575
To detect chirps we used a custom Matlab script to create a spectrogram of the electrical 576 recordings to identify individuals and mark chirp times. Chirps were visually identified as >10Hz 577 abrupt frequency increases.